{
  "name": "AI model release tracker",
  "description": "A dated, source-linked record of every notable model release since 2023, from frontier labs to open-weight and media models. Benchmark scores and prices are left out because they change.",
  "url": "https://theaiengineering.com/releases",
  "publisher": "The AI Engineering",
  "license": "CC BY 4.0",
  "license_url": "https://creativecommons.org/licenses/by/4.0/",
  "citation": "The AI Engineering Editors. \"New AI models: the LLM release tracker.\" The AI Engineering, updated Nov 16, 2026. https://theaiengineering.com/releases",
  "notes": "Dates are the labs' official announcement dates, and every record links to its source.",
  "updated": "2026-11-16",
  "coverage": {
    "from": "2023-01-30",
    "to": "2026-11-16"
  },
  "count": 1167,
  "releases": [
    {
      "slug": "qwen4-max",
      "date": "2026-11-16",
      "model": "Qwen4-Max",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": "MoE: 2.4T total, 71B active, plus 328B N-gram embedding parameters",
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Qwen4-Max is the first model of Alibaba's Qwen4 generation, a sparse mixture-of-experts flagship with native vision and a 1M-token context, aimed at long-running professional, coding, and agent work.",
      "highlights": [
        "API model ID qwen4-max on QwenCloud through the Token Plan or the API, and usable in Qwen Studio, QwenWork, and Qoder at launch",
        "New architecture combining Gated DeltaNet, Qwen Sparse Attention, gated residuals, and N-gram embeddings",
        "Qwen reported a Work360 score of 52.69, up from 47.74 for Qwen3.8-Max, and said it lowered API prices from that model",
        "Qwen said smaller models in the Qwen4 family would follow as open weights"
      ],
      "source_title": "Qwen4-Max: Your New Teammate Has Arrived",
      "source_url": "https://qwen.ai/blog?id=qwen4-max",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/11#qwen4-max"
    },
    {
      "slug": "longcat-video-avatar-1-6",
      "date": "2026-11-15",
      "model": "LongCat-Video-Avatar 1.6",
      "lab": "Meituan",
      "lab_slug": "meituan",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "audio"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "LongCat-Video-Avatar 1.6 is the next version of Meituan LongCat's open-source framework for audio-driven human video generation, with more natural lip movement, more expressive face, head, and body motion, and less error buildup in long videos.",
      "highlights": [
        "Keeps the 1.5 base and adds two task-specific LoRA adapters, plus an optional INT8 base checkpoint",
        "Meituan said it can generate videos longer than 10 minutes while keeping identity and visual quality stable",
        "Ranked second of six systems in Meituan's own human preference study of 33,051 votes",
        "Weights released on Hugging Face and ModelScope under the MIT license, with a technical report"
      ],
      "source_title": "GitHub - meituan-longcat/LongCat-Video",
      "source_url": "https://github.com/meituan-longcat/LongCat-Video",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/11#longcat-video-avatar-1-6"
    },
    {
      "slug": "gemini-4-hydrogen",
      "date": "2026-11-12",
      "model": "Gemini 4 Hydrogen",
      "lab": "Google",
      "lab_slug": "google",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1048576,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms",
        "Consumer app",
        "Enterprise"
      ],
      "summary": "Gemini 4 Hydrogen is the fast, efficient model in Google's Gemini 4 family, built for quick execution of agentic tasks. It launched the same day in the Gemini API, the Gemini app, Search, and Gemini for business.",
      "highlights": [
        "API model ID gemini-4-hydrogen in the Gemini API, Google AI Studio, and Android Studio at launch",
        "1M-token context window, with output of up to 1M tokens",
        "Added an xhigh thinking level above low, medium, and high, which is the default",
        "Designed to pair with Gemini 4 Carbon: in Antigravity's Orchestrator, Carbon coordinates while Hydrogen runs worker and verifier agents"
      ],
      "source_title": "Introducing Gemini 4 Hydrogen",
      "source_url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-hydrogen/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/11#gemini-4-hydrogen"
    },
    {
      "slug": "pplx-decider-v1-3",
      "date": "2026-11-12",
      "model": "pplx-decider-v1.3 and pplx-decider-flash-v1.3",
      "lab": "Perplexity",
      "lab_slug": "perplexity",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "score"
      ],
      "context_window": null,
      "parameters": "27B and 4B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "Perplexity's v1.3 decision models return a probability distribution over predefined answers to yes or no, multiple-choice, and ordered score questions in one forward pass. Decider v1.3 is fine-tuned from Qwen3.8-27B, and Decider Flash v1.3 is distilled from it onto Qwen3.5-4B.",
      "highlights": [
        "Added Locate, which extracts exact text spans from the input with character offsets for entity recognition, PII detection, and extractive QA",
        "Perplexity reported 91.6% mean accuracy across 63 classification tasks for Decider v1.3, 6.54 points above its production classifiers",
        "Accepts images alongside text, as earlier Decider versions did",
        "Weights released on Hugging Face under Apache 2.0, with access through OpenRouter and the Perplexity Decisions API at launch"
      ],
      "source_title": "Perplexity Decider v1.3: Training, eval, and use cases",
      "source_url": "https://www.perplexity.ai/hub/blog/perplexity-decider",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/11#pplx-decider-v1-3"
    },
    {
      "slug": "ernie-5-5-flash-preview",
      "date": "2026-11-11",
      "model": "ERNIE 5.5 Flash Preview",
      "lab": "Baidu",
      "lab_slug": "baidu",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": "MoE: 280B total, 14B active",
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "ERNIE 5.5 Flash Preview is the first post-trained model built on ERNIE 5.5 Flash Base, released as an intermediate checkpoint while Baidu's post-training continued. At launch it was offered through Baidu's Qianfan API, with weights promised later.",
      "highlights": [
        "Intermediate post-training checkpoint of the ERNIE 5.5 Flash Base mixture-of-experts model",
        "280B total and 14B active parameters with a 1M-token context",
        "API on Baidu's Qianfan platform at launch, plus a free OpenRouter listing linked from the announcement",
        "Baidu said at launch that the weights would be open-sourced under Apache 2.0 at a later date"
      ],
      "source_title": "Introducing ERNIE 5.5 Flash Preview",
      "source_url": "https://ernie.baidu.com/release/ernie-5-5-flash-preview/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/11#ernie-5-5-flash-preview"
    },
    {
      "slug": "gemini-decision-1",
      "date": "2026-11-11",
      "model": "Gemini Decision 1",
      "lab": "Google",
      "lab_slug": "google",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "score"
      ],
      "context_window": 262144,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Gemini Decision 1 is a Google model that scores a fixed set of candidate answers in one pass instead of generating text, returning a confidence for each. It is served through the new Gemini Decisions API for tasks such as tool selection, model routing, and policy checks.",
      "highlights": [
        "API model ID gemini-decision-1 on the Decisions API endpoint in the Gemini API and Google AI Studio",
        "Handles yes or no, multiple-choice (2 to 255 options), and ordered rubric questions, up to 1,000 per request, each with a 0 to 1 confidence",
        "Accepts text, images, audio, video, and PDFs, with a 262,144-token input limit",
        "Launched at $0.10 per 1M input tokens with no output charge; Google reported 150 to 250 ms end-to-end latency"
      ],
      "source_title": "Gemini Decisions API: Fast, calibrated multimodal decisions",
      "source_url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemini-decisions-api/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/11#gemini-decision-1"
    },
    {
      "slug": "minimax-h3-image",
      "date": "2026-11-11",
      "model": "MiniMax H3-Image",
      "lab": "MiniMax",
      "lab_slug": "minimax",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "33.1B diffusion transformer",
      "model_license": "MiniMax-H3-Image Community License Agreement",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "MiniMax H3-Image is an open-weights image model built on MiniMax H3 that handles text-to-image generation and image editing in a single checkpoint, generating images at around 16 megapixels.",
      "highlights": [
        "Text-to-image and editing in one checkpoint, with up to 16 reference images",
        "Native output of about 16 MP at every aspect ratio, from 4096x4096 (1:1) to 6272x2688 (21:9)",
        "33.1B-parameter diffusion transformer distilled to 6 sampling steps without classifier-free guidance, sharing its text encoder with MiniMax H3",
        "Weights on Hugging Face, plus access through the MiniMax API and Hailuo AI at launch"
      ],
      "source_title": "MiniMax H3-Image: Bringing 4K Image Generation into the 10-Second Era",
      "source_url": "https://www.minimax.io/blog/minimax-h3-image",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/11#minimax-h3-image"
    },
    {
      "slug": "runway-praxis-1",
      "date": "2026-11-11",
      "model": "Praxis-1",
      "lab": "Runway",
      "lab_slug": "runway",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video",
        "action"
      ],
      "context_window": null,
      "parameters": "27B (mixture of experts)",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Praxis-1 is Runway's first world action model for robot control, a mixture-of-experts diffusion transformer built on Runway's video pretraining that predicts future video and robot actions together.",
      "highlights": [
        "Announced September 30, 2026 with pre-launch access for select partners, then opened to everyone on November 11, 2026",
        "Weights, inference code, and fine-tuning code released under Apache 2.0, including for commercial use",
        "Runway reported 46.2% success across 1,200 episodes on NVIDIA's RoboLab-120 leaderboard",
        "Released on GitHub, on Hugging Face with LeRobot support, and in vLLM-Omni"
      ],
      "source_title": "Runway Praxis-1 Is Now Available: Open-Weight Robotics Model",
      "source_url": "https://runway.com/news/company-news/praxis-1-is-now-available",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/11#runway-praxis-1"
    },
    {
      "slug": "qwen4-max-preview",
      "date": "2026-11-11",
      "model": "Qwen4-Max-Preview",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Qwen4-Max-Preview was a time-limited preview of Alibaba's first Qwen4 flagship, offered through Model Studio's Token Plan subscription ahead of the production Qwen4-Max. It reads text, images, and video and targets coding, office work, frontend design, and agentic tasks.",
      "highlights": [
        "API model ID qwen4-max-preview, offered only through Token Plan Personal and Team, without pay-as-you-go access",
        "Alibaba scheduled the preview from November 11 to November 25, 2026 (UTC+8) and said the model would be retired afterward",
        "1M-token context and up to 393,216 output tokens shared between thinking and the answer",
        "Thinking on by default, with low, medium, and xhigh reasoning effort; followed by Qwen4-Max on November 16, 2026"
      ],
      "source_title": "qwen4-max-preview - Alibaba Cloud Model Studio - Alibaba Cloud Documentation Center",
      "source_url": "https://www.alibabacloud.com/help/en/model-studio/qwen4-max-preview",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/11#qwen4-max-preview"
    },
    {
      "slug": "gemini-robotics-2-5",
      "date": "2026-11-10",
      "model": "Gemini Robotics 2.5",
      "lab": "Google",
      "lab_slug": "google",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "action"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [],
      "summary": "Gemini Robotics 2.5 is Google DeepMind's vision-language-action model for controlling robot arms and hands across different robot bodies, reasoning over several steps before it acts. Google rolled it out to robotics partners instead of offering it publicly.",
      "highlights": [
        "Rolled out at launch to robotics partners, including Apptronik and Mercedes-Benz, with no API access announced",
        "Works across embodiments, including Apptronik's Apollo 2 humanoid and the Franka Duo platform",
        "Google reported 30% to 100% higher success rates and 20% to 30% faster task completion than earlier baselines across 200 tasks",
        "Reached a 96% success rate in a Mercedes-Benz kitting pilot, paired with Gemini Robotics ER for workflow tracking"
      ],
      "source_title": "Gemini Robotics 2.5 accelerates real-world general dexterity",
      "source_url": "https://deepmind.google/blog/gemini-robotics-2-5-accelerates-real-world-physical-intelligence/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/11#gemini-robotics-2-5"
    },
    {
      "slug": "pan-2",
      "date": "2026-11-10",
      "model": "PAN 2",
      "lab": "MBZUAI IFM",
      "lab_slug": "mbzuai-ifm",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": "29B (video diffusion backbone)",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "PAN 2 is an open world model from MBZUAI's Institute of Foundation Models. Given a starting scene and an action described in text, it generates video of the likely outcome, so agents can try out plans before acting.",
      "highlights": [
        "A language model writes a time-stamped description of events, actions, and camera motion, and a video diffusion model trained from scratch by IFM renders it",
        "IFM reported third place overall on the Physics-IQ Verified leaderboard and first among open-weight models",
        "Video model weights and full training and inference code released under Apache 2.0",
        "Model card lists output of up to 20 seconds at 24 fps"
      ],
      "source_title": "PAN 2 Press Release | Institute of Foundation Models",
      "source_url": "https://ifm.ai/pan2/press-release/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/11#pan-2"
    },
    {
      "slug": "elevenlabs-scribe-v3",
      "date": "2026-11-10",
      "model": "Scribe v3 and Scribe v3 Live",
      "lab": "ElevenLabs",
      "lab_slug": "elevenlabs",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Scribe v3 and Scribe v3 Live are ElevenLabs' third generation of speech-to-text models: Scribe v3 transcribes recorded audio, and Scribe v3 Live is a low-latency variant for realtime audio. They succeed Scribe v2 and Scribe v2 Realtime.",
      "highlights": [
        "ElevenLabs reported 27% fewer word errors than Scribe v2 across 102 languages",
        "Scribe v3 Live added speaker diarization and turn detection that uses conversation context",
        "Keyterm prompting took up to 1,000 terms, and Live added a keyterm library for larger vocabularies",
        "At launch, both were in the API as scribe_v3 and scribe_v3_live and were the default in the ElevenLabs app"
      ],
      "source_title": "Introducing Scribe v3 and v3 Live, our most accurate speech to text models",
      "source_url": "https://elevenlabs.io/blog/scribe-v3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/11#elevenlabs-scribe-v3"
    },
    {
      "slug": "claude-fable-5-2",
      "date": "2026-11-09",
      "model": "Claude Fable 5.2",
      "lab": "Anthropic",
      "lab_slug": "anthropic",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Cloud platforms"
      ],
      "summary": "Claude Fable 5.2 succeeds Claude Fable 5.1 for demanding reasoning and long-horizon agentic work. Anthropic says it generates output up to 30% faster than Fable 5.1 and that its cyber and biology safeguards block about 55% fewer requests.",
      "highlights": [
        "API model ID claude-fable-5-2; 1M-token context window and 128K max output tokens",
        "Claude Mythos 5.2, the same underlying model with fewer safeguards, is limited to vetted organizations through Anthropic's Cyber and Life Sciences Verification Programs",
        "New optional fast mode speeds up token generation by up to 2.5 times",
        "Available on all platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure"
      ],
      "source_title": "Introducing Claude Fable 5.2 and Claude Mythos 5.2",
      "source_url": "https://www.anthropic.com/claude-fable-and-mythos-5-2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/11#claude-fable-5-2"
    },
    {
      "slug": "muse-ember",
      "date": "2026-11-09",
      "model": "Muse Ember",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "small",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Cloud platforms"
      ],
      "summary": "Muse Ember is the smallest and most efficient model in Meta's Muse family, designed for complex reasoning execution, with native understanding of text, images, audio, and video.",
      "highlights": [
        "Available at launch on the Meta Model API and in Muse Code",
        "1M-token context window and up to 512K output tokens",
        "The first model trained using Muse Wonder Full",
        "Offered on Google Cloud and Oracle Cloud alongside Muse Spark and Muse Wonder"
      ],
      "source_title": "Introducing Our Smallest, Most Efficient Model: Muse Ember",
      "source_url": "https://research.meta.ai/blog/muse-ember",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/11#muse-ember"
    },
    {
      "slug": "seedream-5-5",
      "date": "2026-11-09",
      "model": "Seedream 5.5",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Seedream 5.5 is ByteDance Seed's next-generation image creation model, adding pixel-level preservation across multi-turn edits, native RGBA output with transparency, and more realistic images with clearer small text.",
      "highlights": [
        "Preserves unedited regions at the pixel level, supporting work on one image across 100 rounds of editing",
        "Generates images directly with an alpha channel, including full and partial transparency",
        "Began rolling out on Jimeng, Doubao, Doubao Work, and Volcano Ark",
        "ByteDance said a Pro version of the Seedream 5.5 series would follow"
      ],
      "source_title": "More Realistic Images, More Stable Editing Across 100 Rounds | Introducing Seedream 5.5",
      "source_url": "https://seed.bytedance.com/en/blog/more-realistic-images-more-stable-editing-across-100-rounds-introducing-seedream-5-5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/11#seedream-5-5"
    },
    {
      "slug": "arete-27b",
      "date": "2026-11-06",
      "model": "Arete-27B",
      "lab": "Nous Research",
      "lab_slug": "nous-research",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "27B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Arete-27B is Nous Research's post-trained version of Alibaba's Qwen3.8-27B, which the lab says it fine-tuned to reinforce Western values and reduce refusals, deflections, and state-framed answers on politically sensitive topics.",
      "highlights": [
        "Supervised fine-tuning only, with no reinforcement learning, preference optimization, or architectural changes",
        "Used Inkling as the teacher and about 52.8 million text-only tokens, trained at 64K context",
        "Weights, the 803-prompt evaluation, and the training data were published alongside the model"
      ],
      "source_title": "Introducing Arete-27B: Qwen3.8-27B Post-Trained To Reinforce Western Values | Nous Research",
      "source_url": "https://nousresearch.com/arete",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/11#arete-27b"
    },
    {
      "slug": "koa-decisions",
      "date": "2026-11-06",
      "model": "Koa Decisions",
      "lab": "Salesforce",
      "lab_slug": "salesforce",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio"
      ],
      "output_modalities": [
        "score"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [],
      "summary": "Koa Decisions is Salesforce's decision model and the successor to HyperClassifier, which routes requests inside Agentforce. It reads a request and a set of user-defined options and returns one choice, a yes or no, or a score in a single step instead of generating text.",
      "highlights": [
        "Built by applying Salesforce's own training techniques to Google's open-weight Gemma",
        "Added image and audio input alongside text",
        "Salesforce measured about 26 ms per Agentforce routing request, against about 1.4 seconds for a general-purpose LLM",
        "At launch Salesforce offered a free Trailhead demo and said an API for developers and support in flows, Apex, and prompt templates would follow"
      ],
      "source_title": "Less Reasoning. More Doing. Meet Koa Decisions.",
      "source_url": "https://www.salesforce.com/news/stories/koa-decisions-salesforce-decision-models/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/11#koa-decisions"
    },
    {
      "slug": "muse-meta-ai-specialized-models",
      "date": "2026-11-06",
      "model": "Specialized models for Muse and Meta AI",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [],
      "output_modalities": [],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Meta introduced specialized versions of its existing models, trained using Muse Wonder, and began rolling them out to people using Muse and Meta AI. The announcement gave the models no separate names.",
      "highlights": [
        "Trained using the Muse Wonder series, which Meta described as its most capable models to date",
        "Brought Muse Wonder's Active Memory and Active Focus to Muse and Meta AI",
        "Meta said the models were developed to multitask, adapt to mid-task clarifications, and ask sharper follow-up questions"
      ],
      "source_title": "Bringing Specialized Models to Muse and Meta AI",
      "source_url": "https://research.meta.ai/blog/specialized-models-for-muse-and-metaai",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/11#muse-meta-ai-specialized-models"
    },
    {
      "slug": "ernie-5-5-flash-base",
      "date": "2026-11-05",
      "model": "ERNIE 5.5 Flash Base",
      "lab": "Baidu",
      "lab_slug": "baidu",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1048576,
      "parameters": "MoE: 280B total, 14B active",
      "model_license": "ERNIE 5.5 Flash Base Model License Agreement v1.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "ERNIE 5.5 Flash Base is the open-weights pre-training checkpoint of Baidu's sparse mixture-of-experts ERNIE 5.5 Flash model, released before any mid- or post-training as a starting point for fine-tuning and reinforcement learning research.",
      "highlights": [
        "Pre-trained from scratch on 33.5T tokens with training context up to 128K; fine-tuning and inference support a 1M-token context",
        "Hybrid attention plus a latent MoE layer with 512 routed experts (top-10) and one shared expert",
        "FP8 and BF16 weights on Hugging Face, with serving support in SGLang and vLLM",
        "Baidu says the post-trained ERNIE 5.5 Flash will follow"
      ],
      "source_title": "ERNIE 5.5 Flash Base",
      "source_url": "https://ernie.baidu.com/release/ernie-5-5-flash-base/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/11#ernie-5-5-flash-base"
    },
    {
      "slug": "gemini-omni-1-2-flash",
      "date": "2026-11-05",
      "model": "Gemini Omni 1.2 Flash",
      "lab": "Google",
      "lab_slug": "google",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "video",
        "audio",
        "text"
      ],
      "context_window": 131072,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms"
      ],
      "summary": "Gemini Omni 1.2 Flash is an update to Google's conversational video generation and editing model that adds native 15-second clips, multi-turn editing and stronger character and subject consistency across reference images.",
      "highlights": [
        "Generally available in the Gemini API and Gemini for business as gemini-omni-1.2-flash",
        "Video duration can be set in one-second steps from 3 to 15 seconds",
        "Requests to gemini-omni-1.1-flash route automatically to 1.2 Flash from Dec 3, 2026"
      ],
      "source_title": "Gemini Omni 1.2 Flash: 15-second video and lower pricing",
      "source_url": "https://blog.google/innovation-and-ai/technology/developers-tools/introducing-gemini-omni-1-2-flash/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/11#gemini-omni-1-2-flash"
    },
    {
      "slug": "gpt-6-cyber",
      "date": "2026-11-05",
      "model": "GPT-6 Cyber",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "coding",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1050000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "GPT-6 Cyber is OpenAI's cybersecurity model, trained to find and validate exploitable vulnerabilities while staying within the scope researchers set. It follows GPT-5.6 Cyber and launched to approved customers of OpenAI's Daybreak Red program.",
      "highlights": [
        "API model ID gpt-6-cyber; initially offered through the API and Codex to a select group of approved Daybreak Red customers",
        "1,050,000-token context window and 128,000 max output tokens per the API model page",
        "OpenAI classified it as Critical in cybersecurity under its Preparedness Framework, up from High for GPT-5.6 Cyber",
        "OpenAI reported using it to develop 948 privilege escalation exploits in Linux and identify 81 previously unknown high-severity vulnerabilities in Firefox"
      ],
      "source_title": "GPT-6 Cyber: Advancing the frontier of aligned cybersecurity agents | OpenAI",
      "source_url": "https://openai.com/index/introducing-gpt-6-cyber/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/11#gpt-6-cyber"
    },
    {
      "slug": "inkling-2-small",
      "date": "2026-11-05",
      "model": "Inkling 2 Small",
      "lab": "Thinking Machines Lab",
      "lab_slug": "thinking-machines-lab",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "MoE: 594B total, 17B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "API"
      ],
      "summary": "Thinking Machines Lab's second model generation: a mixture-of-experts model with image input and variable effort, trained in a single joint reinforcement learning run and launched on Tinker with closed weights.",
      "highlights": [
        "594B total parameters, 17B active",
        "Text and image input; audio is not supported",
        "Launched on Tinker for sampling, fine-tuning and the Playground, with closed weights",
        "Weights released on Hugging Face under Apache 2.0 on November 12, 2026, alongside an MXFP8 checkpoint"
      ],
      "source_title": "Introducing Inkling 2 Small",
      "source_url": "https://thinkingmachines.ai/news/inkling-2-small/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/11#inkling-2-small"
    },
    {
      "slug": "gpt-image-2-6",
      "date": "2026-11-04",
      "model": "GPT Image 2.6 Sunburst and Flare (Images 2.6)",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Images 2.6 is OpenAI's image generation update for ChatGPT, Codex, and the API, adding animated GIF generation, SVG export, and markup-guided editing, with GPT Image 2.6 Sunburst and Flare offered as the API models.",
      "highlights": [
        "API model IDs gpt-image-2.6-sunburst and gpt-image-2.6-flare for image generation and edits",
        "Available globally in ChatGPT and Codex across all subscription tiers on desktop, mobile, and web",
        "OpenAI says Images 2.6 generates images 50% faster than before",
        "API adds an adjustable complexity setting plus markup-guided edits for GPT Image 2.5 and 2.6 models"
      ],
      "source_title": "Introducing Images 2.6 | OpenAI",
      "source_url": "https://openai.com/index/introducing-images-2-6/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/11#gpt-image-2-6"
    },
    {
      "slug": "hy4",
      "date": "2026-11-04",
      "model": "Hy4",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Hy4 is Tencent's flagship Hy model, scaling up pre-training and post-training over Hy4 preview. It adds image understanding and computer use for the first time, with gains in instruction following, long-context work, and long-horizon tasks.",
      "highlights": [
        "First Hy model with vision input and computer use across CLI and GUI environments",
        "Available on Tencent Cloud and OpenRouter, and in CodeBuddy, WorkBuddy, Yuanbao, and ima",
        "Tencent says it plans to open-source the weights along with a system card"
      ],
      "source_title": "Introducing Hy4",
      "source_url": "https://hy.tencent.com/research/hy4",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/11#hy4"
    },
    {
      "slug": "ripple-65b-preview",
      "date": "2026-11-04",
      "model": "Ripple-65B Preview",
      "lab": "Ricursive Intelligence",
      "lab_slug": "ricursive-intelligence",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1048576,
      "parameters": "65B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Ripple-65B Preview is Ricursive Intelligence's open-weights language model, a general-purpose variant of the proprietary model behind its AI chip design platform. It is a dense Transformer trained from scratch, with Ricursive's launch results centered on agentic coding.",
      "highlights": [
        "Pretrained on 12 trillion tokens, with support for up to 1,048,576 tokens of context",
        "Uses a dense Transformer rather than mixture-of-experts, designed to train and serve efficiently on NVIDIA hardware",
        "Released on Hugging Face as an instruction-tuned model in BF16 and NVFP4, plus a BF16 base model",
        "Also hosted at launch by Together AI, GMI Cloud, OpenRouter, and Fireworks"
      ],
      "source_title": "Ripple-65B Preview, a Frontier Open-Weight Model | Ricursive Intelligence",
      "source_url": "https://www.ricursive.com/ripple",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/11#ripple-65b-preview"
    },
    {
      "slug": "swe-2-flash",
      "date": "2026-11-04",
      "model": "SWE-2 Flash",
      "lab": "Cognition",
      "lab_slug": "cognition",
      "type": "coding",
      "weights": "proprietary",
      "input_modalities": [],
      "output_modalities": [],
      "context_window": null,
      "parameters": "MoE: 276B total, 12B active",
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "SWE-2 Flash is Cognition's coding model for everyday engineering work such as automations, code review, feature development, and documentation updates. Cognition built it by adding a mid-training stage and reinforcement learning on top of Thinking Machines' Inkling-Small.",
      "highlights": [
        "Launched in Devin Desktop and CLI, free through November 30, 2026",
        "Cognition reported 3x the speed of its much larger sibling, SWE-2",
        "Mid-training used a few trillion tokens of scientific articles, blogs, code, and other sources",
        "Weights were not released publicly; Cognition said it was building a program to share them securely with trusted industry partners and research labs"
      ],
      "source_title": "Introducing SWE-2 Flash | Cognition",
      "source_url": "https://cognition.com/blog/swe-2-flash",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/11#swe-2-flash"
    },
    {
      "slug": "gpt-rosalind-6",
      "date": "2026-10-30",
      "model": "GPT-Rosalind-6",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [],
      "output_modalities": [],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [],
      "summary": "GPT-Rosalind-6 is OpenAI's frontier reasoning model for biology, drug discovery, and translational medicine. It uses the same weights as GPT-6 Astra and gives fewer refusals on dual-use biological queries, so OpenAI offers it only through trusted access.",
      "highlights": [
        "Offered as GPT-Rosalind-6 Research to qualified scientists, research institutes, companies, and government partners",
        "OpenAI reviews each applicant's proposed uses, security controls, and governance practices before granting access",
        "No added training data or post-training beyond GPT-6 Astra, per the system card addendum",
        "Treated as High capability in the Biological and Chemical domain under OpenAI's Preparedness Framework, below Critical, and still refuses requests for biological weaponization"
      ],
      "source_title": "GPT-Rosalind-6 System Card Addendum",
      "source_url": "https://deploymentsafety.openai.com/gpt-rosalind-6",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#gpt-rosalind-6"
    },
    {
      "slug": "pika-h3-flex",
      "date": "2026-10-29",
      "model": "Pika H3 Flex",
      "lab": "Pika",
      "lab_slug": "pika",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video",
        "audio"
      ],
      "output_modalities": [
        "video",
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "MiniMax H3 Community License Agreement",
      "available_at_launch": [
        "Open weights",
        "API",
        "Consumer app"
      ],
      "summary": "Open-weight omni-generator post-trained from MiniMax H3 to generate synchronized video and stereo audio up to 15 seconds in six denoising steps, accepting up to 12 image, video and audio references.",
      "highlights": [
        "Exports five separate audio stems through the Pika API",
        "Finishes at up to 4K via upscaling",
        "Generator adapters and inference code released on Hugging Face"
      ],
      "source_title": "Introducing Pika H3 Flex",
      "source_url": "https://pika.art/blog/pika-h3-flex",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#pika-h3-flex"
    },
    {
      "slug": "elevenlabs-sfx-v3",
      "date": "2026-10-29",
      "model": "SFX v3",
      "lab": "ElevenLabs",
      "lab_slug": "elevenlabs",
      "type": "music",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Newest ElevenLabs sound effects model, following prompts with multiple sounds more reliably, generating music and vocals more convincingly and producing cleaner, more natural audio than SFX v2.",
      "highlights": [
        "Model ID eleven_text_to_sound_v3",
        "Supports seamless looping sound effects",
        "Default in ElevenCreative Sound Effects, Studio and Flows"
      ],
      "source_title": "Introducing SFX v3",
      "source_url": "https://elevenlabs.io/blog/sound-effects-v3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#elevenlabs-sfx-v3"
    },
    {
      "slug": "gemini-4-carbon",
      "date": "2026-10-28",
      "model": "Gemini 4 Carbon",
      "lab": "Google",
      "lab_slug": "google",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1048576,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [],
      "summary": "Gemini 4 Carbon is Google's frontier model for long-horizon engineering and cybersecurity defense, building on Gemini 4 Argon. Google released it first to a subset of cyber defenders while it continued to iterate on guardrails.",
      "highlights": [
        "Offered first to a subset of cyber defenders, with access phased to additional trusted partners through the Fairwind Program",
        "Became generally available on November 10, 2026 in the Gemini API, Google Cloud, Gemini Enterprise, and Google Antigravity, and began rolling out to Google AI Ultra subscribers in the Gemini app and AI Mode in Search",
        "Google said it was engaged in the U.S. government's voluntary process for pre-release model testing",
        "Built for sustained agentic workflows; Google reported internal runs lasting days on multi-step engineering tasks"
      ],
      "source_title": "Introducing Gemini 4 Carbon",
      "source_url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#gemini-4-carbon"
    },
    {
      "slug": "nemotron-3-5-transcribe",
      "date": "2026-10-28",
      "model": "Nemotron 3.5 Transcribe",
      "lab": "NVIDIA",
      "lab_slug": "nvidia",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "audio",
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 65536,
      "parameters": "34B total, 4.3B active",
      "model_license": "OpenMDW-1.1",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Open-weight offline speech and audio model for transcription in 40 languages, handling recordings up to one hour with a hybrid mixture-of-experts design.",
      "highlights": [
        "Supports speech translation and speaker-attributed transcription",
        "65,536-token context window",
        "Released under the OpenMDW 1.1 license"
      ],
      "source_title": "nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B · Hugging Face",
      "source_url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#nemotron-3-5-transcribe"
    },
    {
      "slug": "d1-2b-omni-2b",
      "date": "2026-10-27",
      "model": "d1-2B and d1-Omni-2B",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video",
        "audio"
      ],
      "output_modalities": [
        "score"
      ],
      "context_window": null,
      "parameters": "d1-2B: 1.92B; d1-Omni-2B: 2.17B",
      "model_license": "LFM Open License v1.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Two open-weight LFM3-based decision models: d1-2B, built on LFM3-VL-2B for text, image, and video inputs, and d1-Omni-2B, built on LFM3-Omni-2B with audio input, adding typed answers such as boxes, values, and estimates.",
      "highlights": [
        "d1-2B replaces d1-3B with about a third fewer parameters",
        "Typed answers: point (boxes), read (text, numbers, dates), estimate, and multi-select",
        "Audio input is d1-Omni-2B only; 128K-token context per model cards",
        "Weights on Hugging Face under the LFM Open License"
      ],
      "source_title": "Introducing LFM3: Every Modality for Specialized Intelligence",
      "source_url": "https://www.liquid.ai/blog/lfm3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#d1-2b-omni-2b"
    },
    {
      "slug": "lfm3-1-7b",
      "date": "2026-10-27",
      "model": "LFM3-1.7B",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 262144,
      "parameters": "1.73B",
      "model_license": "LFM Open License v1.0",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "The text model of Liquid AI's LFM3 generation, built on a new tri-hybrid architecture of gated short convolutions, sliding-window attention, and global attention, pre-trained on 22T tokens and post-trained for local agent runtimes.",
      "highlights": [
        "256K-token context window (262,144 tokens per model card)",
        "24 layers: 13 gated short convolutions, 5 sliding-window attention, 6 global attention",
        "Post-trained for agent runtimes such as Hermes Agent, Pi, and OpenClaw",
        "Open weights on Hugging Face; Liquid API model ID lfm3-1.7b"
      ],
      "source_title": "Introducing LFM3: Every Modality for Specialized Intelligence",
      "source_url": "https://www.liquid.ai/blog/lfm3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#lfm3-1-7b"
    },
    {
      "slug": "lfm3-audio-2b",
      "date": "2026-10-27",
      "model": "LFM3-Audio-2B",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": null,
      "parameters": "2B",
      "model_license": "LFM Open License v1.0",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "A 2B audio model in the LFM3 family with an in-house audio encoder and codec that takes text or speech and replies with text, speech, or both, adding audio understanding, reasoning, and tool calling.",
      "highlights": [
        "Voice chat, transcription, speech synthesis, and speech translation in one model",
        "Speech support expanded to 18 languages",
        "Outputs 24 kHz speech; audio encoder operates at 16 kHz",
        "Liquid API model ID lfm3-audio-2b; open weights on Hugging Face"
      ],
      "source_title": "Introducing LFM3: Every Modality for Specialized Intelligence",
      "source_url": "https://www.liquid.ai/blog/lfm3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#lfm3-audio-2b"
    },
    {
      "slug": "lfm3-omni-2b",
      "date": "2026-10-27",
      "model": "LFM3-Omni-2B",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video",
        "audio"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": null,
      "parameters": "2.17B",
      "model_license": "LFM Open License v1.0",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "Liquid AI's first omni model, combining vision, audio, and language in one LFM3-based model that accepts text, images, video with sound, and speech, and answers in text, speech, or both.",
      "highlights": [
        "Supports spoken visual question answering and audio-visual video understanding",
        "Tool calling from combined visual context and spoken instructions",
        "Liquid API model ID lfm3-omni-2b; open weights on Hugging Face"
      ],
      "source_title": "Introducing LFM3: Every Modality for Specialized Intelligence",
      "source_url": "https://www.liquid.ai/blog/lfm3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#lfm3-omni-2b"
    },
    {
      "slug": "lfm3-vl-2b",
      "date": "2026-10-27",
      "model": "LFM3-VL-2B",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "1.92B",
      "model_license": "LFM Open License v1.0",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "The LFM3 vision-language model, branched from LFM3-1.7B-Base and the first LFM vision model to accept video input, adding step-by-step thinking and spatial reasoning for localizing objects.",
      "highlights": [
        "1.92B parameters: 1.73B language model, 164M vision encoder, 25M projector",
        "Video input sampled at 2 fps per the model card",
        "Liquid API model ID lfm3-vl-2b with a 128,000-token context window",
        "Open weights on Hugging Face"
      ],
      "source_title": "Introducing LFM3: Every Modality for Specialized Intelligence",
      "source_url": "https://www.liquid.ai/blog/lfm3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#lfm3-vl-2b"
    },
    {
      "slug": "mai-image-2-7",
      "date": "2026-10-27",
      "model": "MAI-Image-2.7",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "MAI-Image-2.7 is Microsoft AI's latest image editing model, launched in preview. It improves on MAI-Image-2.6 for restyling, relighting, text and symbol edits, and restoring or enhancing existing images.",
      "highlights": [
        "Launched in preview on Arena, Artificial Analysis and MAI Playground",
        "Starts with image editing; text-to-image generation is listed as coming soon",
        "Succeeds MAI-Image-2.6 with an editing-first launch"
      ],
      "source_title": "MAI Image 2.7 debuts above GPT Image 2.5 Flare on Artificial Analysis and #3 on Arena",
      "source_url": "https://microsoft.ai/news/mai-image-2-7-debut/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#mai-image-2-7"
    },
    {
      "slug": "baidu-unlimited-ocr-2",
      "date": "2026-10-27",
      "model": "Unlimited-OCR 2",
      "lab": "Baidu",
      "lab_slug": "baidu",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "image",
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Unlimited-OCR 2 is Baidu's second-generation open document-parsing vision-language model, adding Sparse Reference Sliding Window Attention so it can parse documents of unbounded length within a finite context window.",
      "highlights": [
        "SR-SWA compresses text-visual attention from ultra-long sequences to a fixed set of working references, such as 512 tokens",
        "Supports multi-page inference that parses all pages of a document in one request",
        "Released with the MarathonDocBench ultra-long document benchmark; also offered on Baidu Cloud",
        "Supports Transformers and SGLang serving"
      ],
      "source_title": "baidu/Unlimited-OCR-2 · Hugging Face",
      "source_url": "https://huggingface.co/baidu/Unlimited-OCR-2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#baidu-unlimited-ocr-2"
    },
    {
      "slug": "gpt-6-1-terra",
      "date": "2026-10-26",
      "model": "GPT-6.1 Terra",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1050000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API",
        "Enterprise"
      ],
      "summary": "GPT-6.1 Terra is a fast, efficient GPT-6.1 model focused on detailed visual reasoning, precise computer use, and clearer writing, which OpenAI describes as a significant step up from GPT-5.6 Terra.",
      "highlights": [
        "API model ID gpt-6.1-terra in the Responses and Chat Completions APIs",
        "In ChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu users; not offered in the Chat tab",
        "1,050,000-token context window and 128,000 max output tokens per the API model page",
        "Tool calling requires the Responses API"
      ],
      "source_title": "Introducing GPT-6.1 Terra | OpenAI",
      "source_url": "https://openai.com/index/introducing-gpt-6-1-terra/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#gpt-6-1-terra"
    },
    {
      "slug": "hy-simulmt",
      "date": "2026-10-26",
      "model": "Hy-SimulMT",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "Consumer app"
      ],
      "summary": "Hy-SimulMT is a real-time simultaneous translation model for bidirectional Chinese and English speech. It translates continuously as audio arrives and, paired with a configured TTS service, supports speech output with real-time voice cloning.",
      "highlights": [
        "Released as Hy-SimulMT-30B-A3B plus an Interleave variant that outputs source text and translation",
        "FP8 and GGUF checkpoints published alongside the full weights",
        "Also usable in the Hy Translate app"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/Hy-SimulMT",
      "source_url": "https://github.com/Tencent-Hunyuan/Hy-SimulMT",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#hy-simulmt"
    },
    {
      "slug": "kimi-k3-1-pro",
      "date": "2026-10-24",
      "model": "Kimi K3.1 Pro",
      "lab": "Moonshot AI",
      "lab_slug": "moonshot-ai",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1048576,
      "parameters": "MoE: 2.8T total, 104B active",
      "model_license": "Kimi K3.1 Pro License",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "Kimi K3.1 Pro is the first model in Moonshot AI's K3.1 series, a post-trained version of Kimi K3 built with substantially larger post-training compute. It improves coding, 3D creation, financial research, and other agentic work.",
      "highlights": [
        "Released with Kimi K3.1 Pro Highspeed, which runs about 2 to 3 times faster",
        "Available on Kimi.com, Kimi Work, Kimi Code, and the Kimi API as kimi-k3.1-pro",
        "Weights published on Hugging Face under the Kimi K3.1 Pro License"
      ],
      "source_title": "Introducing Kimi K3.1 Pro",
      "source_url": "https://www.kimi.com/en/blog/kimi-k3-1-pro",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#kimi-k3-1-pro"
    },
    {
      "slug": "ideogram-vector-1-0",
      "date": "2026-10-22",
      "model": "Ideogram Vector 1.0",
      "lab": "Ideogram",
      "lab_slug": "ideogram",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Ideogram's first vector model, generating clean, editable SVGs with smooth gradients and sharp typography from a text prompt, or converting an existing image into a scalable vector.",
      "highlights": [
        "Each design is one SVG editable in Illustrator, Figma or code",
        "Swapping the palette updates the gradients to match",
        "Live in Ideogram, the API and partner platforms"
      ],
      "source_title": "Ideogram on X: Today, we're introducing Ideogram's first vector model.",
      "source_url": "https://x.com/ideogram_ai/status/2113300317584720259",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#ideogram-vector-1-0"
    },
    {
      "slug": "hy-nomas",
      "date": "2026-10-21",
      "model": "Hy-Nomas",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Hy-Nomas is an open interactive world generation model that produces explorable video worlds controlled by multi-turn text prompts or camera inputs. The release includes training and inference code, LoRA fine-tuning, and an interactive web demo.",
      "highlights": [
        "Runs on a single RTX 4090 with a low-memory inference configuration",
        "Default output resolution of 720x1280",
        "Image references for initializing a world were added to the demo on October 31, 2026"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/Hy-Nomas: Hy-Nomas: Towards Free-form Interactive World Generation",
      "source_url": "https://github.com/Tencent-Hunyuan/Hy-Nomas",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#hy-nomas"
    },
    {
      "slug": "cohere-rerank-5",
      "date": "2026-10-21",
      "model": "Rerank 5",
      "lab": "Cohere",
      "lab_slug": "cohere",
      "type": "reranker",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "score"
      ],
      "context_window": 32000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Reranking family in Pro and Fast variants that adds Context Focus, which returns the most relevant passages inside each document, and calibrated relevance scores.",
      "highlights": [
        "Model IDs rerank-v5.0-pro and rerank-v5.0-fast",
        "Supports 100+ languages",
        "Added to OpenRouter on November 4, 2026"
      ],
      "source_title": "Cohere Rerank 5 | More Relevant Context for RAG and Agents",
      "source_url": "https://cohere.com/blog/rerank-5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#cohere-rerank-5"
    },
    {
      "slug": "eleven-realtime-v1",
      "date": "2026-10-20",
      "model": "Eleven Realtime v1 Mini and Max",
      "lab": "ElevenLabs",
      "lab_slug": "elevenlabs",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Enterprise"
      ],
      "summary": "New class of speech-to-speech voice agent models built on a cascaded architecture, each a complete agent configuration; Mini targets natural, low-cost conversation and Max handles complex multi-step tasks.",
      "highlights": [
        "Supports more than 90 languages",
        "Compatible with OpenAI Realtime API events and transport",
        "Rolled out to every workspace by November 2, 2026"
      ],
      "source_title": "Eleven Realtime",
      "source_url": "https://elevenlabs.io/blog/realtime",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#eleven-realtime-v1"
    },
    {
      "slug": "granite-speech-5-0",
      "date": "2026-10-20",
      "model": "Granite Speech 5.0",
      "lab": "IBM",
      "lab_slug": "ibm",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "2B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Speech-language model for multilingual automatic speech recognition and bidirectional speech translation in English, French, German, Spanish, Portuguese and Japanese, trained on 390,000 hours of audio.",
      "highlights": [
        "Adds language ID prompting and larger keyword lists up to 2,000",
        "Companion 470M TurboCTC model went multilingual on October 21, 2026",
        "Released under Apache 2.0"
      ],
      "source_title": "ibm-granite/granite-speech-5.0-2b · Hugging Face",
      "source_url": "https://huggingface.co/ibm-granite/granite-speech-5.0-2b",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#granite-speech-5-0"
    },
    {
      "slug": "voxtral-3",
      "date": "2026-10-20",
      "model": "Voxtral 3",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Enterprise"
      ],
      "summary": "Speech model family of Voxtral Small Transcribe 3, Voxtral Mini Transcribe 3, Voxtral Mini Transcribe Realtime 3 and Voxtral TTS 3, launched in public preview with a Voice Agents SDK.",
      "highlights": [
        "Batch transcription covers 50+ languages with diarization",
        "TTS 3 supports 35+ languages and 50+ dialects",
        "Available via the EU-hosted API or self-deployment"
      ],
      "source_title": "Voxtral 3: Speech-to-Text, TTS & Voice Agents | Mistral",
      "source_url": "https://mistral.ai/news/voxtral-3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#voxtral-3"
    },
    {
      "slug": "minimax-m3-1-flash",
      "date": "2026-10-19",
      "model": "MiniMax M3.1 Flash",
      "lab": "MiniMax",
      "lab_slug": "minimax",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": "MoE: about 428B total, 23B active",
      "model_license": "MiniMax Community License",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "MiniMax M3.1 Flash is an open-weight multimodal mixture-of-experts model for coding, agentic workflows, computer use, and cybersecurity. It builds on MiniMax M3 with sparse attention throughout, quantization-aware training, and DSpark speculative decoding.",
      "highlights": [
        "Accepts text, image, and video input with a context window of up to 1M tokens",
        "Thinking depth is adjustable through an effort parameter in the API",
        "An early version ran anonymously under the name Space Bunny before launch",
        "M Plan requests for M2.5 and M2.7 are routed to M3.1 Flash from October 19, 2026"
      ],
      "source_title": "MiniMax M3.1 Flash: Redefining Agent Performance and Cost",
      "source_url": "https://www.minimax.io/blog/minimax-m31f",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#minimax-m3-1-flash"
    },
    {
      "slug": "deepseek-v4-1-flash-1017",
      "date": "2026-10-17",
      "model": "DeepSeek-V4.1-Flash-1017",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": "MoE: 552B backbone, 8B active (prefill), 16B active (decode)",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "DeepSeek-V4.1-Flash-1017 is a minor update to DeepSeek-V4.1-Flash that improves reasoning, coding, agentic, and visual results while shortening outputs and task runtime. It replaced the previous version on the DeepSeek web, app, and API.",
      "highlights": [
        "Served through the deepseek-flash API model name with unchanged pricing and calling method",
        "Trained with reinforcement learning with self-feedback for end-to-end project development",
        "Weights on Hugging Face and ModelScope under the MIT License"
      ],
      "source_title": "DeepSeek-V4.1-Flash-1017: Wish-Driven Programming Made Real",
      "source_url": "https://api-docs.deepseek.com/news/news261017",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#deepseek-v4-1-flash-1017"
    },
    {
      "slug": "beam-1-0",
      "date": "2026-10-16",
      "model": "Beam 1.0",
      "lab": "Reflection AI",
      "lab_slug": "reflection-ai",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": "MoE: 501B total, 23B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "API",
        "Cloud platforms"
      ],
      "summary": "Beam 1.0 is Reflection AI's first open-weights model, a sparse mixture-of-experts language model built for advanced coding, reasoning, and long-horizon agentic workloads.",
      "highlights": [
        "501B total parameters with 23B active and a 1M-token context window, per the model card",
        "Weights on Hugging Face under Apache 2.0 in BF16, FP8, and NVFP4, with a speculative decoding drafter model",
        "Served through Reflection's own API platform, free to start during its public beta, and by hosts including AWS, Google Cloud, Microsoft Foundry, OpenRouter, and Ollama",
        "Dell and HPE offer it for on-premises, sovereign, and air-gapped deployments",
        "Developed predominantly for English"
      ],
      "source_title": "Beam is now available — Reflection",
      "source_url": "https://reflection.ai/blog/beam-is-available",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#beam-1-0"
    },
    {
      "slug": "d1-ultra",
      "date": "2026-10-15",
      "model": "d1-ultra",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "score"
      ],
      "context_window": 32768,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "A hosted multimodal decision model that Liquid AI built on a larger open-weight base than d1 with Liquid-specific customizations, trained to produce better-calibrated probabilities for harder text and image decisions.",
      "highlights": [
        "Liquid API model ID d1-ultra on the decisions endpoint",
        "Returns Noul, Choice, and Score answers without generating tokens",
        "32,768-token context window per Liquid API docs",
        "Later listed on OpenRouter and Vercel AI Gateway per Liquid docs"
      ],
      "source_title": "d1-ultra: Introducing the World’s Best Decision Model",
      "source_url": "https://www.liquid.ai/blog/d1-ultra",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#d1-ultra"
    },
    {
      "slug": "flux-3-image-klein",
      "date": "2026-10-15",
      "model": "FLUX 3 Image Klein",
      "lab": "Black Forest Labs",
      "lab_slug": "black-forest-labs",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "5.6B",
      "model_license": "FLUX Kommunity License",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "5.6B-parameter open-weight version of FLUX 3 Image for text-to-image, editing and multi-reference generation with bounding-box control, released as a distilled model and a base model for fine-tuning.",
      "highlights": [
        "Up to 4K output and up to 10 reference images",
        "FP8, INT8 and NVFP4 builds developed with NVIDIA",
        "Runs on GeForce RTX GPUs with as little as 8GB VRAM using the INT8 build"
      ],
      "source_title": "FLUX 3 Image Klein",
      "source_url": "https://bfl.ai/blog/flux-3-image-klein",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#flux-3-image-klein"
    },
    {
      "slug": "gpt-live-1-1",
      "date": "2026-10-15",
      "model": "GPT-Live-1.1 and GPT-Live-1.1 Quip",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "GPT-Live-1.1 is an updated version of OpenAI's full-duplex GPT-Live voice model for ChatGPT Voice, rolled out with GPT-Live-1.1 Quip, a mini version, and bringing more natural accents across languages.",
      "highlights": [
        "Rolled out in ChatGPT Voice alongside GPT-Live-1.1 Quip, a mini version of the model",
        "Uses a wider range of short spoken phrases to signal when it is searching",
        "New Live voices for French, German, Hindi, Japanese, Korean, and Mexican Spanish, plus Bubbie in English"
      ],
      "source_title": "ChatGPT release notes | OpenAI Help Center",
      "source_url": "https://help.openai.com/en/articles/6825453-chatgpt-release-notes",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#gpt-live-1-1"
    },
    {
      "slug": "jais-3",
      "date": "2026-10-15",
      "model": "Jais 3",
      "lab": "MBZUAI IFM",
      "lab_slug": "mbzuai-ifm",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [],
      "output_modalities": [],
      "context_window": null,
      "parameters": "0.9B to 375B; largest is MoE with 375B total, 23B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Jais 3 is a family of six open Arabic-centric models from 0.9B to 375B parameters, built by MBZUAI's Institute of Foundation Models on its K2 Horizon models and trained to work as agents in Modern Standard Arabic, dialects, and Arabizi.",
      "highlights": [
        "Sizes: 0.9B, 3.7B, 7B, and 32B dense models, a 36B-A4B mixture-of-experts model with MoVA attention, and a 375B-A23B mixture-of-experts model",
        "512K-token context window, or 128K for the 0.9B model",
        "Each model starts from the post-trained K2 Horizon model of the same size",
        "Released with checkpoints from every training stage, a sample of the training data, training logs, and evaluation code; models and code are under Apache 2.0"
      ],
      "source_title": "Jais 3: Introducing Arabic-centric Agentic AI",
      "source_url": "https://ifm.ai/blog/jais3/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#jais-3"
    },
    {
      "slug": "kling-4-0",
      "date": "2026-10-15",
      "model": "Kling 4.0",
      "lab": "Kuaishou",
      "lab_slug": "kuaishou",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video",
        "audio"
      ],
      "output_modalities": [
        "video",
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Kling 4.0 is Kuaishou's next-generation video generation model, focused on steadier large-scale motion and complex camera work, native single-pass generation of up to 30 seconds, and expanded reference-based control.",
      "highlights": [
        "Native single-pass video generation of up to 30 seconds",
        "Omni Reference supports up to 15 reference assets and up to 10 keyframes",
        "Supports up to 10-bit HDR output",
        "Kuaishou said API access for enterprises and developers would follow"
      ],
      "source_title": "Kling AI Officially Launches Next-Generation Kling 4.0, Raising the Bar for Professional-Grade Video Models",
      "source_url": "https://ir.kuaishou.com/news-releases/news-release-details/kling-ai-officially-launches-next-generation-kling-40-raising",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#kling-4-0"
    },
    {
      "slug": "muse-spark-open",
      "date": "2026-10-15",
      "model": "Muse Spark Open",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video",
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": "MoE: 816B total, 42B active",
      "model_license": "Muse Model License",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "Muse Spark Open is the open-weights member of Meta Superintelligence Labs' Muse family, a natively multimodal mixture-of-experts model built for coding and agentic work with a 1M-token context window.",
      "highlights": [
        "816B total parameters with 42B active per token",
        "Model card lists text, image, video and audio input with text output",
        "Weights on Hugging Face, Amazon SageMaker AI, Google Cloud Model Garden and Microsoft Foundry",
        "Hosted by providers including Baseten, Fireworks AI, Groq and Together AI"
      ],
      "source_title": "Introducing Muse Spark Open, The Most Capable US Open-Weight Model to Date",
      "source_url": "https://research.meta.ai/blog/muse-spark-open",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#muse-spark-open"
    },
    {
      "slug": "nemotron-3-5-bolt",
      "date": "2026-10-15",
      "model": "Nemotron 3.5 Bolt",
      "lab": "NVIDIA",
      "lab_slug": "nvidia",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "74B total, 9B active; E4B 8B with 4B effective",
      "model_license": "OpenMDW-1.1",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Compressed Nemotron 3.5 models sized for memory budgets, including a 74B mixture-of-experts model targeting a 24GB footprint and an E4B model targeting 4GB.",
      "highlights": [
        "A budget-conditioned router selects expert counts per layer",
        "The E4B model followed on October 23, 2026 with a 64K-token window",
        "Released under the OpenMDW 1.1 license"
      ],
      "source_title": "nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB · Hugging Face",
      "source_url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#nemotron-3-5-bolt"
    },
    {
      "slug": "nemotron-3-5-super-vl",
      "date": "2026-10-15",
      "model": "Nemotron 3.5 Super VL",
      "lab": "NVIDIA",
      "lab_slug": "nvidia",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": "MoE: 120B total, 12B active",
      "model_license": "OpenMDW-1.1",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Open multimodal Nemotron 3.5 model that combines Mamba-2, mixture-of-experts and attention layers with a C-RADIO vision encoder and a vision-language connector.",
      "highlights": [
        "Supports context lengths up to 1M tokens",
        "Base and post-trained checkpoints released",
        "Released under the OpenMDW 1.1 license"
      ],
      "source_title": "nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16 · Hugging Face",
      "source_url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#nemotron-3-5-super-vl"
    },
    {
      "slug": "timestral",
      "date": "2026-10-15",
      "model": "Timestral",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "time series"
      ],
      "output_modalities": [
        "time series"
      ],
      "context_window": null,
      "parameters": "5M, 28M, 123M and 209M",
      "model_license": "Modified MIT License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Mistral's first family of time-series foundation models, built for zero-shot probabilistic and multivariate forecasting with support for past-only and known-future covariates across four model sizes.",
      "highlights": [
        "Four sizes from 5M to 209M parameters",
        "Runs on CPUs without a GPU",
        "Supports contexts up to 32K timesteps"
      ],
      "source_title": "Timestral: Mistral's time-series foundation models | Mistral",
      "source_url": "https://mistral.ai/news/timestral",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#timestral"
    },
    {
      "slug": "grok-voice-3-0",
      "date": "2026-10-14",
      "model": "Grok Voice 3.0",
      "lab": "xAI",
      "lab_slug": "xai",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio",
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Grok Voice 3.0 is SpaceXSI's speech-to-speech model for real-time voice agents. It succeeds Grok Voice 2.0 in the Speech to Speech API at the same price, and the grok-voice-latest alias was upgraded automatically.",
      "highlights": [
        "API model ID: grok-voice-3.0",
        "grok-voice-latest now points to grok-voice-3.0",
        "SpaceXSI reports response latency under one second",
        "Grok Voice 2.0 was deprecated the same day"
      ],
      "source_title": "Introducing Grok Voice 3.0 | SpaceXSI",
      "source_url": "https://x.ai/news/grok-voice-3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#grok-voice-3-0"
    },
    {
      "slug": "jev-1-14",
      "date": "2026-10-14",
      "model": "Jev 1.14",
      "lab": "TypeSafe AI",
      "lab_slug": "typesafe-ai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "score"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "An update to TypeSafe AI's Jev with small improvements across the board, the largest in multilingual support, plus opt-in image input and long context and a new box answer type.",
      "highlights": [
        "API model ID jev-1.14.0",
        "Image input and long context sat behind opt-in danger-zone headers, which TypeSafe said were not yet ready for production use",
        "New box answer type, available only on jev-1.14.0",
        "Offered in early access, like the original Jev"
      ],
      "source_title": "Jev: For the builders AND dreamers",
      "source_url": "https://typesafe.ai/blog/jev-1-14",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#jev-1-14"
    },
    {
      "slug": "lfm2-5-24b-a2b",
      "date": "2026-10-13",
      "model": "LFM2.5-24B-A2B",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 131072,
      "parameters": "MoE: 24.0B total, 2.45B active",
      "model_license": "LFM Open License v1.0",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "The largest and final LFM2.5 model, a reasoning-only mixture-of-experts model with 24B total and about 2B active parameters, extending LFM2-24B-A2B with a 131K context window, larger vocabulary, and agentic reinforcement learning.",
      "highlights": [
        "Context extended from 32,768 to 131,072 tokens; vocabulary doubled from 65K to 128K",
        "Pre-training scaled from 17T to 33T tokens",
        "Agentic RL inside agent runtimes such as OpenClaw, Hermes Agent, Pi, and OpenCode",
        "Base and post-trained weights on Hugging Face; also on Liquid Playground and Liquid API"
      ],
      "source_title": "LFM2.5-24B-A2B: Built for Fast On-Device Agents",
      "source_url": "https://www.liquid.ai/blog/lfm2-5-24b-a2b",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#lfm2-5-24b-a2b"
    },
    {
      "slug": "muse-wonder",
      "date": "2026-10-12",
      "model": "Muse Wonder and Muse Wonder Full",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video",
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1048576,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Muse Wonder and Muse Wonder Full open the Muse Wonder series, Meta's most capable models, aimed at personal and professional agentic tasks with Active Memory and Active Focus features.",
      "highlights": [
        "Muse Wonder is available through the Meta Model API and Muse Code",
        "Muse Wonder Full is available to Muse Code subscribers only",
        "Meta Model API docs list a 1,048,576-token context window for Muse Wonder",
        "Muse Wonder is slated to power the Muse personal agent, first for subscribers"
      ],
      "source_title": "Muse Wonder: A Frontier Model For Personal & Professional AI Agents",
      "source_url": "https://research.meta.ai/blog/muse-wonder",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#muse-wonder"
    },
    {
      "slug": "claude-haiku-5-5",
      "date": "2026-10-07",
      "model": "Claude Haiku 5.5",
      "lab": "Anthropic",
      "lab_slug": "anthropic",
      "type": "small",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Cloud platforms"
      ],
      "summary": "Claude Haiku 5.5 is the small, high-volume model in Anthropic's Claude 5.5 family, aimed at summaries, classification, subagent work, and latency-sensitive tasks such as customer support and browser use.",
      "highlights": [
        "API model ID claude-haiku-5-5",
        "1M-token context window and 128K max output tokens",
        "First Haiku-class model with an adjustable effort setting",
        "Available in the Claude apps and on Amazon Web Services, Google Cloud, and Microsoft Azure"
      ],
      "source_title": "Introducing Claude Haiku 5.5",
      "source_url": "https://www.anthropic.com/claude-haiku-5-5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#claude-haiku-5-5"
    },
    {
      "slug": "d1-3b-omni-600m",
      "date": "2026-10-07",
      "model": "d1-3B and d1-omni-600M",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "audio"
      ],
      "output_modalities": [
        "score"
      ],
      "context_window": null,
      "parameters": "d1-3B: 3.12B; d1-omni-600M: 587M",
      "model_license": "LFM Open License v1.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "The first open-weight d1 decision models: d1-3B, trained from LFM2.5-VL-3B for text and image inputs, and the experimental d1-omni-600M, built on LFM2.5-Encoder-350M with added vision and audio encoders.",
      "highlights": [
        "Answers questions in one forward pass without generating tokens",
        "d1-omni-600M accepts text with images or text with audio",
        "Context: 32,768 tokens (d1-3B) and 16,384 tokens (d1-omni-600M) per model cards",
        "Available on Hugging Face with day-one llama.cpp support"
      ],
      "source_title": "Open d1: Edge decision models for text, vision, and audio",
      "source_url": "https://www.liquid.ai/blog/d1-open",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#d1-3b-omni-600m"
    },
    {
      "slug": "gpt-6-in-chatgpt",
      "date": "2026-10-07",
      "model": "GPT-6 in ChatGPT (GPT-6 Instant, with Intelligent UI)",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "Enterprise"
      ],
      "summary": "GPT-6 in ChatGPT brings the GPT-6 generation to the Chat tab for all tiers, introducing Intelligent UI, which composes answers from text, visuals, and interactive elements, and letting ChatGPT begin answering while it keeps thinking.",
      "highlights": [
        "Powered by GPT-6 Sol for Plus, Pro, Business, and Enterprise, and GPT-6 Luna for Free and Go",
        "Rolled out to paid tiers on October 7, 2026, expanding to Free and Go tiers the next day",
        "Intelligent UI responses can include graphics, tappable buttons, forms, charts, and interactive tools",
        "Models powering ChatGPT Work and Codex did not change with this release"
      ],
      "source_title": "GPT-6 and Intelligent UI for everyone | OpenAI",
      "source_url": "https://openai.com/index/gpt-6-for-everyone/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#gpt-6-in-chatgpt"
    },
    {
      "slug": "pplx-embed-v2-late",
      "date": "2026-10-07",
      "model": "pplx-embed-v2-late-0.6b and pplx-embed-v2-late-9b",
      "lab": "Perplexity",
      "lab_slug": "perplexity",
      "type": "embedding",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "embedding"
      ],
      "context_window": null,
      "parameters": "0.6b: 594M total; 9b: 7.4B active",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "A family of late-interaction (ColBERT-style) embedding models in 0.6B and 9B sizes that output 128-dimensional vectors per token for text and image retrieval, including rendered PDF pages, in a shared embedding space.",
      "highlights": [
        "Scores query-document pairs with MaxSim over 128-dimensional token embeddings",
        "The 0.6B model can query an index built with the 9B model through the shared embedding space",
        "Built on Qwen3.5 with bidirectional attention; the 0.6B model is pruned from Qwen3.5-0.8B",
        "Weights on Hugging Face under the MIT license"
      ],
      "source_title": "Multimodal embeddings beyond a single vector",
      "source_url": "https://www.perplexity.ai/hub/blog/multimodal-embeddings-beyond-a-single-vector",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#pplx-embed-v2-late"
    },
    {
      "slug": "embeddinggemma-2",
      "date": "2026-10-06",
      "model": "EmbeddingGemma 2",
      "lab": "Google",
      "lab_slug": "google",
      "type": "embedding",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "embedding"
      ],
      "context_window": 8192,
      "parameters": "740M (270M text backbone)",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "EmbeddingGemma 2 is a lightweight open multimodal embedding model built on the Gemma 4 architecture. It maps text, code, images, audio and video into one shared embedding space, succeeding the text-only EmbeddingGemma.",
      "highlights": [
        "8K-token context window, four times larger than EmbeddingGemma 1",
        "768-dimension output with Matryoshka truncation to 512, 256 or 128",
        "Text-only use needs 270M parameters; vision (170M) and audio (300M) encoders are optional",
        "Weights on Hugging Face and Kaggle under Apache 2.0"
      ],
      "source_title": "EmbeddingGemma 2: an open, lightweight multimodal embedding model",
      "source_url": "https://blog.google/innovation-and-ai/technology/developers-tools/embeddinggemma-2/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#embeddinggemma-2"
    },
    {
      "slug": "mistral-large-4",
      "date": "2026-10-06",
      "model": "Mistral Large 4",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": "MoE: 1T total, 52B active",
      "model_license": "Mistral Ecosystem Open-Weight License",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Mistral's largest model, a natively multimodal mixture-of-experts model nicknamed Le Chonk, launched as an API preview and aimed at agentic coding, cybersecurity and knowledge work.",
      "highlights": [
        "Improved weights released October 29, 2026 under a modified MIT license",
        "Three reasoning modes balance cost and performance",
        "Self-hostable with vLLM or SGLang"
      ],
      "source_title": "Introducing Mistral Large 4 | Mistral",
      "source_url": "https://mistral.ai/news/mistral-large-4",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#mistral-large-4"
    },
    {
      "slug": "nano-banana-2-1",
      "date": "2026-10-06",
      "model": "Nano Banana 2.1",
      "lab": "Google",
      "lab_slug": "google",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text",
        "image"
      ],
      "context_window": 131072,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Nano Banana 2.1 is an update to Nano Banana 2 that keeps Flash-level speed while improving visual quality, prompt adherence, multi-turn character consistency and text rendering for image generation and editing.",
      "highlights": [
        "Model ID gemini-nano-banana-2.1, released as generally available in the Gemini API",
        "Adds wide and panoramic aspect ratios (1:4, 4:1, 1:8 and 8:1) at 1K, 2K and 4K",
        "gemini-3.1-flash-image (Nano Banana 2) was deprecated the same day"
      ],
      "source_title": "Release notes - Gemini API | Google AI Studio",
      "source_url": "https://ai.google.dev/gemini-api/docs/changelog",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#nano-banana-2-1"
    },
    {
      "slug": "amazon-nova-2-5-sonic",
      "date": "2026-10-05",
      "model": "Amazon Nova 2.5 Sonic",
      "lab": "Amazon",
      "lab_slug": "amazon",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": 256000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Amazon Nova 2.5 Sonic is Amazon's speech-to-speech model for real-time voice agents, succeeding Nova 2 Sonic with improved reasoning, instruction following and tool-calling accuracy plus lower response latency.",
      "highlights": [
        "Generally available in Amazon Bedrock at launch",
        "256K token context window with up to 64K output tokens",
        "Expressive voices in seven languages, with asynchronous tool calling and mixed voice and text sessions"
      ],
      "source_title": "Announcing Amazon Nova 2.5 Sonic with improved reasoning for voice agents",
      "source_url": "https://aws.amazon.com/about-aws/whats-new/2026/10/amazon-nova-2.5-Sonic/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#amazon-nova-2-5-sonic"
    },
    {
      "slug": "d1-decision-model",
      "date": "2026-10-05",
      "model": "d1",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "score"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Liquid AI's first decision model, which reads text and images with one or more questions in a single forward pass and returns probabilities for each answer instead of generating tokens; this release adds image input.",
      "highlights": [
        "Liquid API model ID d1, served through the decisions endpoint",
        "Question types: Noul (yes/no), Choice, and Score",
        "Also on OpenRouter and Vercel AI Gateway, text only at launch",
        "Follows an experimental text-only d1 release the prior week"
      ],
      "source_title": "Introducing d1: The most capable decision model, now with vision",
      "source_url": "https://www.liquid.ai/blog/d1-decision-model",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#d1-decision-model"
    },
    {
      "slug": "astabrief",
      "date": "2026-10-02",
      "model": "AstaBrief 8B",
      "lab": "Ai2",
      "lab_slug": "ai2",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "8B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Open-weights model that turns a research question and retrieved literature excerpts into a cited scientific report, powering Fast mode in Asta's report generation.",
      "highlights": [
        "Post-trained from Qwen3-8B",
        "Training data open-sourced with the model",
        "Released under Apache 2.0"
      ],
      "source_title": "Open-sourcing AstaBrief, the fast report-generation model in Asta | Ai2",
      "source_url": "https://allenai.org/blog/astabrief",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#astabrief"
    },
    {
      "slug": "flux-3-image",
      "date": "2026-10-01",
      "model": "FLUX 3 Image",
      "lab": "Black Forest Labs",
      "lab_slug": "black-forest-labs",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Image generation and editing endpoint built on FLUX 3 that uses one prompt for both modes, with bounding boxes for layout control, pixel-exact local edits and up to ten reference images.",
      "highlights": [
        "Output from 768px previews up to 4K",
        "Bounding boxes are added as a JSON list in the prompt",
        "Custom output sizes and better single-reference edits added October 22, 2026"
      ],
      "source_title": "Release Notes - Black Forest Labs",
      "source_url": "https://docs.bfl.ai/release-notes",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#flux-3-image"
    },
    {
      "slug": "mai-transcribe-2-streaming",
      "date": "2026-10-01",
      "model": "MAI-Transcribe-2-Streaming",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Cloud platforms"
      ],
      "summary": "MAI-Transcribe-2-Streaming is Microsoft AI's first streaming speech-to-text model, delivering low-latency, real-time transcripts for conversational voice agents in 60 languages with automatic, continuous language detection.",
      "highlights": [
        "Real-time transcription in 60 languages",
        "Automatic, continuous language detection",
        "Available through Microsoft Foundry, MAI Playground, Vercel and Azure Voice Live"
      ],
      "source_title": "Our first streaming transcription model debuts at no. 1 on Artificial Analysis",
      "source_url": "https://microsoft.ai/news/our-first-streaming-transcription-model/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#mai-transcribe-2-streaming"
    },
    {
      "slug": "mai-voice-2-1",
      "date": "2026-10-01",
      "model": "MAI-Voice-2.1",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Cloud platforms"
      ],
      "summary": "MAI-Voice-2.1 is Microsoft AI's text-to-speech model announced alongside MAI-Transcribe-2-Streaming for conversational voice agents. Microsoft calls it its strongest multilingual text-to-speech model, now covering 23 languages and 26 locales.",
      "highlights": [
        "Supports 23 languages and 26 locales",
        "Available through Microsoft Foundry, MAI Playground, OpenRouter, Vercel and Azure Voice Live",
        "Released with a faster MAI-Voice-2.1-Flash variant"
      ],
      "source_title": "Our first streaming transcription model debuts at no. 1 on Artificial Analysis",
      "source_url": "https://microsoft.ai/news/our-first-streaming-transcription-model/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#mai-voice-2-1"
    },
    {
      "slug": "mai-voice-2-1-flash",
      "date": "2026-10-01",
      "model": "MAI-Voice-2.1-Flash",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Cloud platforms"
      ],
      "summary": "MAI-Voice-2.1-Flash is a faster, high-volume variant of Microsoft AI's MAI-Voice-2.1 text-to-speech model, built for latency-sensitive voice agent workloads while supporting the same languages and cross-language speakers.",
      "highlights": [
        "Supports the same 23 languages as MAI-Voice-2.1",
        "Available through Microsoft Foundry, MAI Playground, OpenRouter, Vercel and Azure Voice Live",
        "Announced alongside MAI-Transcribe-2-Streaming"
      ],
      "source_title": "Our first streaming transcription model debuts at no. 1 on Artificial Analysis",
      "source_url": "https://microsoft.ai/news/our-first-streaming-transcription-model/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#mai-voice-2-1-flash"
    },
    {
      "slug": "suno-speech",
      "date": "2026-10-01",
      "model": "Suno Speech (beta)",
      "lab": "Suno",
      "lab_slug": "suno",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Beta audio model in Suno that generates spoken audio set to original background music from text, with the voice and musical style described in the prompt.",
      "highlights": [
        "Generates voice and music together",
        "Built into the Suno app",
        "Released as a beta"
      ],
      "source_title": "Introducing Speech (beta)",
      "source_url": "https://suno.com/blog/introducing-speech-beta",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/10#suno-speech"
    },
    {
      "slug": "cohere-embed-5",
      "date": "2026-09-30",
      "model": "Embed 5",
      "lab": "Cohere",
      "lab_slug": "cohere",
      "type": "embedding",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "embedding"
      ],
      "context_window": 128000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms",
        "Enterprise"
      ],
      "summary": "Cohere's fifth-generation embedding family, in Pro and Fast variants that share an embedding space, for retrieval over visually rich documents, filings, code and multilingual data.",
      "highlights": [
        "Model IDs embed-v5.0-pro and embed-v5.0-fast",
        "Matryoshka dimensions from 256 to 2048",
        "Available on Microsoft Foundry and Amazon SageMaker"
      ],
      "source_title": "Embed 5: Frontier Embedding Models for Enterprise | Cohere",
      "source_url": "https://cohere.com/blog/embed-5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#cohere-embed-5"
    },
    {
      "slug": "gemini-4-argon",
      "date": "2026-09-30",
      "model": "Gemini 4 Argon",
      "lab": "Google",
      "lab_slug": "google",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1048576,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms",
        "Consumer app",
        "Enterprise"
      ],
      "summary": "Gemini 4 Argon is Google's frontier reasoning model for complex coding, agentic, math, science and enterprise work, with a unified 1 million token input and output limit in the Gemini API.",
      "highlights": [
        "Initially rolled out to trusted cyber defenders through the Fairwind Program",
        "Generally available on Oct 13, 2026 in the Gemini API, Google Cloud, Google Antigravity, the Gemini app and AI Mode in Search",
        "Output limit raised to 1 million tokens from the previous 64K via a long decode continuation protocol"
      ],
      "source_title": "Introducing Gemini 4 Argon",
      "source_url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#gemini-4-argon"
    },
    {
      "slug": "ideogram-4-5",
      "date": "2026-09-30",
      "model": "Ideogram 4.5",
      "lab": "Ideogram",
      "lab_slug": "ideogram",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Ideogram image editing model built for multi-turn editing, changing only what a prompt asks so repeated edits do not build up artifacts, pixel shifts or color changes.",
      "highlights": [
        "Four quality modes, all at native 2K resolution",
        "Edits stylized text in place or translates it",
        "Ideogram said open weights are coming soon"
      ],
      "source_title": "Ideogram on X: Introducing Ideogram 4.5, the most precise edit model.",
      "source_url": "https://x.com/ideogram_ai/status/2105327223431737780",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#ideogram-4-5"
    },
    {
      "slug": "pplx-embed-v2-context",
      "date": "2026-09-30",
      "model": "pplx-embed-v2-context-9b-preview",
      "lab": "Perplexity",
      "lab_slug": "perplexity",
      "type": "embedding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "embedding"
      ],
      "context_window": null,
      "parameters": "9B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "A preview contextual embedding model that embeds document chunks with their surrounding context, trained by distilling chunk relevance from Perplexity's context compression model to retrieve supporting context as well as answer chunks.",
      "highlights": [
        "Returns one INT8 embedding per chunk, with 1024- or 2048-dimensional (MRL) vectors",
        "Non-preview pplx-embed-v2-context-0.6b and -9b (text and image) followed on Hugging Face in October 2026",
        "Weights on Hugging Face under the MIT license"
      ],
      "source_title": "Contextual embedding beyond the gold passage",
      "source_url": "https://www.perplexity.ai/hub/blog/contextual-embedding-beyond-the-gold-passage",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#pplx-embed-v2-context"
    },
    {
      "slug": "gpt-6-1-sol",
      "date": "2026-09-29",
      "model": "GPT-6.1 Sol",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1050000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API",
        "Enterprise"
      ],
      "summary": "GPT-6.1 Sol is an upgrade to GPT-6 Sol for agentic coding, computer use, and professional work, which OpenAI positions as delivering near GPT-6 Astra intelligence at substantially lower cost.",
      "highlights": [
        "API model ID gpt-6.1-sol in the Responses and Chat Completions APIs; multi-agent delegation in beta",
        "In ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu users; not in the Chat tab at launch",
        "1,050,000-token context window and 128,000 max output tokens per the API model page",
        "Ultrafast mode for gpt-6.1-sol reached the Responses API on October 8, 2026"
      ],
      "source_title": "Introducing GPT-6.1 Sol | OpenAI",
      "source_url": "https://openai.com/index/introducing-gpt-6-1-sol/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#gpt-6-1-sol"
    },
    {
      "slug": "claude-sonnet-5-5",
      "date": "2026-09-28",
      "model": "Claude Sonnet 5.5",
      "lab": "Anthropic",
      "lab_slug": "anthropic",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Cloud platforms"
      ],
      "summary": "Claude Sonnet 5.5 is the second model in Anthropic's Claude 5.5 family, positioned as a faster, lower-cost complement to Claude Opus 5.5 for well-scoped everyday tasks, bug fixing, and documents, slides, and spreadsheets.",
      "highlights": [
        "API model ID claude-sonnet-5-5; 1M-token context window and 128K max output tokens",
        "First Sonnet model to launch with cyber safeguards and fallbacks",
        "New between_tools thinking setting replaces disabling thinking",
        "Default effort is Medium in Claude Code and the Claude apps, High on the Claude Platform"
      ],
      "source_title": "Introducing Claude Sonnet 5.5",
      "source_url": "https://www.anthropic.com/claude-sonnet-5-5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#claude-sonnet-5-5"
    },
    {
      "slug": "eleven-v4",
      "date": "2026-09-28",
      "model": "Eleven v4 and Eleven v4 Turbo",
      "lab": "ElevenLabs",
      "lab_slug": "elevenlabs",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "ElevenLabs' most emotive text-to-speech model, built on a new architecture to interpret tone, pacing and character, with a low-latency Eleven v4 Turbo variant for agents.",
      "highlights": [
        "Supports 90+ languages",
        "Eleven v4 Turbo has a median inference latency of about 100ms",
        "Served through the Text to Dialogue API and WebSocket"
      ],
      "source_title": "Introducing Eleven v4, our most emotive model",
      "source_url": "https://elevenlabs.io/blog/eleven-v4",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#eleven-v4"
    },
    {
      "slug": "longcat-2-5",
      "date": "2026-09-25",
      "model": "LongCat-2.5",
      "lab": "Meituan",
      "lab_slug": "meituan",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "MoE: 1.6T total, about 48B active",
      "model_license": "MIT",
      "available_at_launch": [
        "API"
      ],
      "summary": "LongCat-2.5 adds image understanding to Meituan's LongCat line and targets agentic coding, combining visual understanding with long-horizon engineering tasks. A preview launched on the LongCat API in September, followed by the full release in November 2026.",
      "highlights": [
        "LongCat-2.5-Preview became available on the LongCat API on September 25, 2026",
        "LongCat-2.5 became available on November 2, 2026",
        "Supports visual coding, creative design, and 3D tasks",
        "Works with Claude Code, Codex, Hermes, and other coding agents",
        "Weights published on Hugging Face under the MIT license in November 2026"
      ],
      "source_title": "Change Log | LongCat API Docs",
      "source_url": "https://longcat.chat/platform/docs/change-log",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#longcat-2-5"
    },
    {
      "slug": "flux-3-action",
      "date": "2026-09-23",
      "model": "FLUX 3 Action",
      "lab": "Black Forest Labs",
      "lab_slug": "black-forest-labs",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "video",
        "action"
      ],
      "context_window": null,
      "parameters": "7B",
      "model_license": "FLUX Kommunity License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Open-weight 7B world action model derived from the FLUX 3 multimodal backbone that predicts robot actions jointly with video, released with weights, code and a report on its fine-tuning recipe.",
      "highlights": [
        "7B parameters, pretrained on video, image and audio data",
        "Weights released under the FLUX Kommunity License",
        "Targets robot control, with gaming explored as a testbed"
      ],
      "source_title": "FLUX 3 Action: A 7B World Action Model for Robot Control",
      "source_url": "https://bfl.ai/blog/flux-3-action",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#flux-3-action"
    },
    {
      "slug": "gemini-3-8-flash-tts",
      "date": "2026-09-23",
      "model": "Gemini 3.8 Flash TTS",
      "lab": "Google",
      "lab_slug": "google",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": 8192,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Enterprise"
      ],
      "summary": "Gemini 3.8 Flash TTS is Google's flagship creative text-to-speech model, built for studio-grade voice fidelity, nuanced acting and long-form multi-turn stability, with voices designed from natural language prompts.",
      "highlights": [
        "Voice design across more than 100 languages and dialects, plus voice replication from a 30-second sample with consent verification",
        "Launched with the Gemini API Voices endpoint and an extended voice library",
        "Generated audio carries SynthID watermarking"
      ],
      "source_title": "Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS",
      "source_url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-8-text-to-speech/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#gemini-3-8-flash-tts"
    },
    {
      "slug": "gemini-3-8-flash-lite-tts",
      "date": "2026-09-23",
      "model": "Gemini 3.8 Flash-Lite TTS",
      "lab": "Google",
      "lab_slug": "google",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": 8192,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Enterprise"
      ],
      "summary": "Gemini 3.8 Flash-Lite TTS is a fast, cost-efficient text-to-speech model for high-throughput production use and real-time voice agent pipelines, replacing the Gemini 3.1 Flash TTS preview.",
      "highlights": [
        "Model ID gemini-3.8-flash-lite-tts, generally available in the Gemini API from Sep 22, 2026",
        "Released alongside Gemini 3.8 Flash TTS and the Gemini API Voices endpoint",
        "Input token limit of 8,192 per the Gemini API model page"
      ],
      "source_title": "Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS",
      "source_url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-8-text-to-speech/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#gemini-3-8-flash-lite-tts"
    },
    {
      "slug": "recraft-v4-1-flash",
      "date": "2026-09-23",
      "model": "Recraft V4.1 Flash",
      "lab": "Recraft",
      "lab_slug": "recraft",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Fast variant of Recraft V4.1 that generates an image in about 1.3 seconds from prompt while keeping the Recraft look across photography, portraits, complex scenes and early concepts.",
      "highlights": [
        "About 1.3 seconds from prompt to image",
        "Available in Recraft Studio"
      ],
      "source_title": "Meet Recraft V4.1 Flash: The Fastest Image Model on the Market",
      "source_url": "https://www.recraft.ai/blog/meet-recraft-v4-1-flash",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#recraft-v4-1-flash"
    },
    {
      "slug": "claude-opus-5-5",
      "date": "2026-09-22",
      "model": "Claude Opus 5.5",
      "lab": "Anthropic",
      "lab_slug": "anthropic",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Cloud platforms"
      ],
      "summary": "Claude Opus 5.5 is the first model in Anthropic's Claude 5.5 family, built for long-running agentic coding and knowledge work. It ships with safeguards similar to Claude Fable 5.1 and requires less compute to serve than Opus 5.",
      "highlights": [
        "API model ID claude-opus-5-5; 1M-token context window and 128K max output tokens",
        "Adaptive thinking is always on and cannot be disabled",
        "Preserved thinking applies to API accounts created on or after August 31, 2026",
        "Available on Amazon Web Services, Google Cloud, and Microsoft Azure; zero data retention available"
      ],
      "source_title": "Introducing Claude Opus 5.5",
      "source_url": "https://www.anthropic.com/claude-opus-5-5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#claude-opus-5-5"
    },
    {
      "slug": "gpt-6-sol",
      "date": "2026-09-22",
      "model": "GPT-6 Sol and GPT-6 Luna",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1050000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API",
        "Enterprise"
      ],
      "summary": "GPT-6 Sol and GPT-6 Luna extend the GPT-6 family with lower-cost tiers trained with methods similar to GPT-6 Astra, with Sol aimed at complex coding and agentic workflows and Luna at focused, high-volume tasks.",
      "highlights": [
        "API model IDs gpt-6-sol and gpt-6-luna, with text and image input and text output",
        "In ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu; Free and Go get Luna in the desktop app",
        "Both list a 1,050,000-token context window on their API model pages",
        "Not available in the ChatGPT Chat tab at launch"
      ],
      "source_title": "Introducing GPT-6 Sol and Luna | OpenAI",
      "source_url": "https://openai.com/index/introducing-gpt-6-sol-and-luna/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#gpt-6-sol"
    },
    {
      "slug": "hy-image-3-5-preview",
      "date": "2026-09-22",
      "model": "Hy Image 3.5 preview",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Hy Image 3.5 preview is Tencent's image generation model building on the Hy Image 3.0 series, with improved text rendering and layout, realism and style range, and editing consistency for everyday and professional visual work.",
      "highlights": [
        "Supports text-to-image and image-to-image with up to five reference images",
        "Multi-turn creation and continuous editing, with output up to 2K resolution",
        "Available through the Tencent Cloud API, Hy AI Studio, Yuanbao, and ima"
      ],
      "source_title": "Hy Image3.5 preview 发布：为专业创作提供高性价比模型",
      "source_url": "https://hy.tencent.com/research/hy-image35-preview",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#hy-image-3-5-preview"
    },
    {
      "slug": "mimo-v2-6",
      "date": "2026-09-22",
      "model": "MiMo-V2.6-Pro and MiMo-V2.6-Flash",
      "lab": "Xiaomi",
      "lab_slug": "xiaomi",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video",
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": "MoE: 1.02T total, 42B active (Pro); 309B total, 15B active (Flash)",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "MiMo-V2.6 is Xiaomi's open-weight series of two natively omnimodal models, MiMo-V2.6-Pro and MiMo-V2.6-Flash, post-trained with substantially scaled reinforcement learning on verifiable coding, agent, visual, and cyber tasks.",
      "highlights": [
        "1M-token context with text, image, video, and audio input",
        "Weights released under MIT along with the technical report, training environments, and RL code",
        "Launched with MiMo-V2.6-Pro-UltraSpeed, a faster serving option on the API",
        "Both models completed a live-streamed RL run of under six days"
      ],
      "source_title": "MiMo-V2.6 | Xiaomi",
      "source_url": "https://mimo.xiaomi.com/mimo-v2-6/article.html",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#mimo-v2-6"
    },
    {
      "slug": "elevenlabs-scribe-v2-medical",
      "date": "2026-09-22",
      "model": "Scribe v2 Medical",
      "lab": "ElevenLabs",
      "lab_slug": "elevenlabs",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Speech-to-text model fine-tuned from Scribe v2 for clinical audio, handling drug names, dosages and medical terminology, with the same features, languages and API as Scribe v2.",
      "highlights": [
        "Model ID scribe_v2_medical",
        "Generally available on ElevenAPI",
        "Built for protected health information workflows"
      ],
      "source_title": "Scribe v2 Medical is generally available",
      "source_url": "https://elevenlabs.io/blog/scribe-v2-medical-generally-available",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#elevenlabs-scribe-v2-medical"
    },
    {
      "slug": "grok-4-7",
      "date": "2026-09-21",
      "model": "Grok 4.7",
      "lab": "xAI",
      "lab_slug": "xai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 500000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms"
      ],
      "summary": "Grok 4.7 is SpaceXSI's frontier model for coding, agentic tasks, and knowledge work. It uses a new, larger base model than Grok 4.6 and a longer reinforcement learning run weighted toward multi-hour tasks.",
      "highlights": [
        "API model ID: grok-4.7, with a 500k-token context window",
        "Reasoning effort settings: low, medium, high (default), and xhigh",
        "Available at launch in Cursor and Grok Build; a fast variant doubles output speed",
        "Trained to work natively inside Grok Bot"
      ],
      "source_title": "Introducing Grok 4.7 | SpaceXSI",
      "source_url": "https://x.ai/news/grok-4-7",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#grok-4-7"
    },
    {
      "slug": "qwen-image-2-1",
      "date": "2026-09-20",
      "model": "Qwen-Image-2.1",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "7B (visual generation component)",
      "model_license": "Qwen Research License Agreement",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "Qwen-Image-2.1 is an open-weights Qwen image model that unifies text-to-image generation and image editing in a single model. It is designed to balance generation quality, inference efficiency, and cost, and natively supports transparent images.",
      "highlights": [
        "Visual generation component has 32 single-stream DiT layers and 7B parameters",
        "Qwen-Image-2.1-Turbo, which needs only 8 denoising steps, followed on October 9, 2026",
        "Pro and Turbo APIs went live on Alibaba Cloud Model Studio on October 9, 2026"
      ],
      "source_title": "Qwen-Image-2.1: Compact, Efficient, and Unified Image Creation",
      "source_url": "https://qwen.ai/blog?id=qwen-image-2.1",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#qwen-image-2-1"
    },
    {
      "slug": "step-5-preview",
      "date": "2026-09-20",
      "model": "Step 5 Preview",
      "lab": "StepFun",
      "lab_slug": "stepfun",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": "MoE: 600B total, 27B active",
      "model_license": "StepFun Model License 1.0",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "Step 5 Preview is StepFun's flagship model for agentic work, aimed at software engineering and professional knowledge work with particular strength in finance. It launched through StepFun products and API, with open weights following on October 15, 2026.",
      "highlights": [
        "Sparse mixture-of-experts model with 600B total and 27B active parameters",
        "1M-token context window; accepts text, image, and video input",
        "API model ID step-5-preview with low, medium, and high reasoning effort",
        "BF16 and FP8 weights on Hugging Face under the StepFun Model License 1.0"
      ],
      "source_title": "阶跃发布旗舰模型 Step 5 Preview：向前一步，智能效率的新一代“帕累托前沿”",
      "source_url": "https://www.stepfun.com/news/step-5-preview",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#step-5-preview"
    },
    {
      "slug": "grok-voice-transcribe-2-0",
      "date": "2026-09-18",
      "model": "Grok Voice Transcribe 2.0",
      "lab": "xAI",
      "lab_slug": "xai",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Grok Voice Transcribe 2.0 is SpaceXSI's speech-to-text model, built on the audio foundation model behind Grok Voice. It replaces Grok Voice Transcribe 1.0 as the default Speech to Text model, with its largest gains in multilingual accuracy.",
      "highlights": [
        "API model ID: grok-voice-transcribe-2.0, the default Speech to Text model",
        "Batch and streaming modes, word-level timestamps, and speaker diarization",
        "Detects language automatically and follows mid-recording language switches",
        "grok-voice-transcribe-1.0 requests route to 2.0 from October 2, 2026"
      ],
      "source_title": "Introducing Grok Voice Transcribe 2.0 | SpaceXSI",
      "source_url": "https://x.ai/news/grok-voice-transcribe-2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#grok-voice-transcribe-2-0"
    },
    {
      "slug": "qwen3-8-omni-flash",
      "date": "2026-09-18",
      "model": "Qwen3.8-Omni-Flash",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": 1000000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Qwen3.8-Omni-Flash is Qwen's omni-modal model for real-time audio-visual interaction and agentic delivery. It accepts text, image, audio, and video input with a 1M-token context window.",
      "highlights": [
        "Qwen reports text performance comparable to a text-only model of the same size",
        "Upgraded video captioning lets users set subject, time range, detail, and format"
      ],
      "source_title": "Qwen3.8-Omni-Flash: Omni Senses. Agentic Delivery.",
      "source_url": "https://qwen.ai/blog?id=qwen3.8-omni-flash",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#qwen3-8-omni-flash"
    },
    {
      "slug": "lfm2-longevity",
      "date": "2026-09-17",
      "model": "LFM2-1.2B-Longevity and LFM2-2.6B-Longevity",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32768,
      "parameters": "1.2B; 2.6B",
      "model_license": "LFM Open License v1.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Domain-adapted LFM2 models fine-tuned on aging-biology data spanning clinical records, DNA methylation, transcriptomics, proteomics, and genetics, released with the LongevityBench benchmark and an accompanying Cell paper.",
      "highlights": [
        "Fine-tuned from LFM2-1.2B and LFM2-2.6B for three epochs at a 32k-token context",
        "LongevityBench covers 17 tasks and 25,457 prompts across five biodata domains",
        "Weights available on Hugging Face"
      ],
      "source_title": "Setting the Frontier of Aging Biology with Liquid Foundation Models",
      "source_url": "https://www.liquid.ai/blog/longevitybench",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#lfm2-longevity"
    },
    {
      "slug": "gemini-3-8-live",
      "date": "2026-09-15",
      "model": "Gemini 3.8 Live",
      "lab": "Google",
      "lab_slug": "google",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": 131072,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Enterprise"
      ],
      "summary": "Gemini 3.8 Live is a native audio model for real-time conversation in the Live API, built for sub-second latency, near real-time visual context and background tool use without interrupting the conversation.",
      "highlights": [
        "Detects and switches between 97 supported languages mid-conversation",
        "Available in the Gemini API, Google Workspace and the Gemini app",
        "Gemini Enterprise added a Live Avatar feature on Sep 24, 2026"
      ],
      "source_title": "Gemini 3.8 Live & Gemini 3.8 Live Extended Thinking",
      "source_url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-8-live-gemini-3-8-live-extended-thinking/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#gemini-3-8-live"
    },
    {
      "slug": "gemini-3-8-live-extended-thinking",
      "date": "2026-09-15",
      "model": "Gemini 3.8 Live Extended Thinking",
      "lab": "Google",
      "lab_slug": "google",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": 131072,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Gemini 3.8 Live Extended Thinking is a speech-to-speech Live API model specialized for complex reasoning, multi-step planning and long-horizon state tracking during real-time voice conversations.",
      "highlights": [
        "Model ID gemini-3.8-live-extended-thinking with configurable thinking levels from MINIMAL to HIGH",
        "Supports asynchronous function calling",
        "Input token limit of 131,072 per the Gemini API model page"
      ],
      "source_title": "Gemini 3.8 Live & Gemini 3.8 Live Extended Thinking",
      "source_url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-8-live-gemini-3-8-live-extended-thinking/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#gemini-3-8-live-extended-thinking"
    },
    {
      "slug": "jev",
      "date": "2026-09-15",
      "model": "Jev",
      "lab": "TypeSafe AI",
      "lab_slug": "typesafe-ai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "score"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "TypeSafe AI's first public model and the first of what it calls System One models: it reads unstructured data and program state and returns type-safe structured values with calibrated probabilities instead of generating text.",
      "highlights": [
        "No string generation: answers come back as typed values with confidence scores",
        "Trained with what TypeSafe calls Reinforcement Learning for Calibrated Decisions (RLCD)",
        "TypeSafe reported 70–500 ms end-to-end response times on its own benchmarks",
        "Launched in early access on TypeSafe's hosted API at $0.042 per million input tokens, with output tokens free"
      ],
      "source_title": "Introducing System One Models & Jev",
      "source_url": "https://typesafe.ai/blog/introducing-system-one-models-and-jev",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#jev"
    },
    {
      "slug": "stepaudio-3",
      "date": "2026-09-15",
      "model": "StepAudio 3",
      "lab": "StepFun",
      "lab_slug": "stepfun",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "StepAudio 3 is StepFun's speech and audio model family, comprising StepAudio 3 Realtime, ASR, TTS, Gen, and Music for full-duplex voice interaction, speech recognition, speech synthesis, multi-element audio generation, and music creation.",
      "highlights": [
        "StepAudio 3 Realtime supports native full-duplex conversation with asynchronous tool calls",
        "StepAudio 3 Gen produces voices, sound effects, ambience, and background music from text descriptions and reference audio",
        "StepAudio 3 Music supports song creation, a cappella accompaniment, and covers, with ABC notation control",
        "All five models launched on the StepFun open platform"
      ],
      "source_title": "阶跃发布语音大模型StepAudio 3，拿下多个全球第一！",
      "source_url": "https://www.stepfun.com/news/stepaudio-3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#stepaudio-3"
    },
    {
      "slug": "eleven-music-v2-5",
      "date": "2026-09-11",
      "model": "Music v2.5",
      "lab": "ElevenLabs",
      "lab_slug": "elevenlabs",
      "type": "music",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Most advanced ElevenLabs music model, producing more layered and complex songs with more natural-sounding instruments, and the default for prompted and reference generation in ElevenMusic.",
      "highlights": [
        "Model ID music_v2_5",
        "Default model in ElevenMusic",
        "Improved quality and prompt adherence over Music v2"
      ],
      "source_title": "Music v2.5",
      "source_url": "https://elevenlabs.io/blog/music-v2-5-model",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#eleven-music-v2-5"
    },
    {
      "slug": "deepseek-v4-1-flash",
      "date": "2026-09-10",
      "model": "DeepSeek-V4.1-Flash",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": "MoE: 552B backbone, 8B active (prefill), 16B active (decode)",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "DeepSeek-V4.1-Flash is the smallest model in DeepSeek's new architecture family, a multimodal mixture-of-experts model with native image understanding. It uses a Causal Encoder-Decoder design that activates fewer parameters for input than for output.",
      "highlights": [
        "Pre-trained from scratch on a 45T-token multimodal corpus",
        "Global KV cache about 4 times smaller than DeepSeek-V4-Flash",
        "Replaced V4-Flash and V4-Flash-Vision-Exp; API model name deepseek-flash",
        "Supports a continuously controllable reasoning effort from 1 to 100"
      ],
      "source_title": "DeepSeek-V4.1-Flash: Smarter, Faster, More Efficient",
      "source_url": "https://api-docs.deepseek.com/news/news260910",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#deepseek-v4-1-flash"
    },
    {
      "slug": "north-small-translate",
      "date": "2026-09-10",
      "model": "North Small Translate",
      "lab": "Cohere",
      "lab_slug": "cohere",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 16000,
      "parameters": "MoE: 218B total, 25B active",
      "model_license": "CC-BY-NC-4.0",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "Open-weights mixture-of-experts model from Cohere built specifically for machine translation across more than 50 languages and locale variants, with weights offered for non-commercial use.",
      "highlights": [
        "Available through the Chat V2 API",
        "Weights in W4A16, FP8 and BF16 for non-commercial use",
        "Runs on two H100s or one B200 in W4A16"
      ],
      "source_title": "Introducing North Small Translate | Cohere",
      "source_url": "https://cohere.com/blog/north-small-translate",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#north-small-translate"
    },
    {
      "slug": "hunyuan-auk",
      "date": "2026-09-09",
      "model": "AuK",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": "1.5B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "AuK is an open 1.5B foundation model from Tencent Hunyuan for speech generation and editing, trained on millions of hours of audio. It supports zero-shot and instruction-based TTS, content and acoustic editing, and paralinguistic editing.",
      "highlights": [
        "Released together with AuK-Flash, a distilled variant",
        "Weights on Hugging Face, with demo spaces on Hugging Face and ModelScope",
        "Day-zero serving support in SGLang-Omni"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/AuK: AuK: An Open-Source Foundational Model for Speech Generation and Editing",
      "source_url": "https://github.com/Tencent-Hunyuan/AuK",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#hunyuan-auk"
    },
    {
      "slug": "suno-v6",
      "date": "2026-09-09",
      "model": "Suno v6 (v6, v6-wild, v6-mini)",
      "lab": "Suno",
      "lab_slug": "suno",
      "type": "music",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "audio",
        "image",
        "video"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "New generation of Suno music models developed with industry partners including Warner Music Group, BMG and Believe, offered as flagship v6, exploratory v6-wild and fast, free v6-mini.",
      "highlights": [
        "Up to 8 minutes of music in a single generation",
        "v6 and v6-wild for Pro and Premier; v6-mini for all plans",
        "Previous models to be retired as v6 rolls out"
      ],
      "source_title": "Introducing v6",
      "source_url": "https://suno.com/blog/introducing-v6",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#suno-v6"
    },
    {
      "slug": "gpt-image-2-5",
      "date": "2026-09-08",
      "model": "GPT Image 2.5 Sunburst and Flare (ChatGPT Images 2.5)",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "ChatGPT Images 2.5 is OpenAI's updated image model with sharper detail, more precise multi-turn editing, and lower latency, offered in the API as GPT Image 2.5 Flare for everyday generation and Sunburst for precision editing.",
      "highlights": [
        "API model IDs gpt-image-2.5-flare and gpt-image-2.5-sunburst in the Image API and Responses image tool",
        "Available to ChatGPT, ChatGPT Work, and Codex users across all tiers on desktop, mobile, and web",
        "OpenAI reports up to 50% lower image generation latency than Images 2.0",
        "Both API models support new xhigh and max quality settings"
      ],
      "source_title": "Introducing ChatGPT Images 2.5 | OpenAI",
      "source_url": "https://openai.com/index/introducing-chatgpt-images-2-5/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#gpt-image-2-5"
    },
    {
      "slug": "mai-image-2-6-flash",
      "date": "2026-09-04",
      "model": "MAI-Image-2.6-Flash",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "MAI-Image-2.6-Flash is a faster variant of Microsoft AI's MAI-Image-2.6, built for latency-sensitive, high-throughput image generation and editing workloads. It launched in Microsoft Foundry alongside developer access to MAI-Image-2.6.",
      "highlights": [
        "Supports multi-image reference editing, web grounding and dynamic aspect ratios",
        "Available to developers in Microsoft Foundry",
        "Generally available in Foundry per an October 27, 2026 update"
      ],
      "source_title": "Pushing the quality-cost frontier with MAI-Image-2.6",
      "source_url": "https://microsoft.ai/news/pushing-the-quality-cost-frontier-with-mai-image-2-6/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#mai-image-2-6-flash"
    },
    {
      "slug": "gpt-6-astra",
      "date": "2026-09-03",
      "model": "GPT-6 Astra",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1050000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API",
        "Cloud platforms",
        "Enterprise"
      ],
      "summary": "GPT-6 Astra is the first model of OpenAI's GPT-6 generation, positioned for the hardest end-to-end work across reasoning, coding, computer use, research, and document creation, with added misalignment monitoring during agent tasks.",
      "highlights": [
        "API model ID gpt-6-astra; tool calling requires the Responses API and the none reasoning effort is unsupported",
        "Rolled out to a limited set of organizations, then ChatGPT Plus, Pro, Business, Enterprise, Azure, and Bedrock",
        "GPT-6 Astra Pro offered to Pro, Business, and Enterprise users",
        "1,050,000-token context window and 128,000 max output tokens per the API model page"
      ],
      "source_title": "GPT-6 Astra: A new generation of intelligence | OpenAI",
      "source_url": "https://openai.com/index/gpt-6-astra/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#gpt-6-astra"
    },
    {
      "slug": "k2-horizon",
      "date": "2026-09-03",
      "model": "K2 Horizon",
      "lab": "MBZUAI IFM",
      "lab_slug": "mbzuai-ifm",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [],
      "output_modalities": [],
      "context_window": null,
      "parameters": "0.9B to 375B; largest is MoE with 375B total, 23B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "K2 Horizon is a family of six fully open foundation models from 0.9B to 375B parameters, released by MBZUAI's Institute of Foundation Models with weights, code, training data, and methodology for reasoning, math, coding, and agentic tasks.",
      "highlights": [
        "Sizes: 0.9B, 3.7B, 7B, and 32B dense models, a 36B-A4B sparse model using Mixture of Value Attention (MoVA), and a 375B-A23B flagship",
        "A diffusion distillation technique generates blocks of tokens in parallel, which IFM says speeds up generation by roughly 3x",
        "The models share one architecture, vocabulary, and training method, with a smaller vocabulary for the 0.9B model",
        "Weights on Hugging Face with vLLM and SGLang support, and an API through inference partners including Compass, Cerebras, and Nebius",
        "Models and code released under Apache 2.0"
      ],
      "source_title": "K2 Horizon Press Release | Institute of Foundation Models",
      "source_url": "https://ifm.ai/k2/press-release/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#k2-horizon"
    },
    {
      "slug": "mai-transcribe-2",
      "date": "2026-09-03",
      "model": "MAI-Transcribe-2",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Cloud platforms"
      ],
      "summary": "MAI-Transcribe-2 is Microsoft AI's batch speech-to-text model, adding speaker diarization, word-level timestamps and configurable transcription styles, with accuracy across 60 languages, up from 43 in MAI-Transcribe-1.5.",
      "highlights": [
        "Covers 60 languages",
        "Adds speaker diarization, word-level timestamps, keyword biasing and automatic language identification",
        "Available through Microsoft Foundry, MAI Playground and OpenRouter"
      ],
      "source_title": "MAI-Transcribe-2 is the fastest, most accurate and cheapest speech recognition model in the world",
      "source_url": "https://microsoft.ai/news/mai-transcribe-2-is-the-fastest-most-accurate-and-cheapest-speech-recognition-model-in-the-world/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#mai-transcribe-2"
    },
    {
      "slug": "gemini-3-8-flash",
      "date": "2026-09-02",
      "model": "Gemini 3.8 Flash",
      "lab": "Google",
      "lab_slug": "google",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1048576,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms",
        "Consumer app"
      ],
      "summary": "Gemini 3.8 Flash is Google's most intelligent Flash model at release, with significant gains over 3.7 Flash in long-horizon software engineering, autonomous agents and multi-step reasoning in specialized domains.",
      "highlights": [
        "Generally available in the Gemini API via Google AI Studio and Android Studio, in Google Antigravity and in Stitch",
        "Applies extra reasoning steps and iterative tool calls on complex tasks, with lower effort levels to limit token use",
        "Announced with Gemini 3.8 Flash Cyber, which is limited to trusted defenders through the Fairwind Program"
      ],
      "source_title": "Introducing Gemini 3.8 Flash and 3.8 Flash Cyber",
      "source_url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/3-8-flash-and-3-8-flash-cyber/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#gemini-3-8-flash"
    },
    {
      "slug": "muse-spark-1-3",
      "date": "2026-09-02",
      "model": "Muse Spark 1.3",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video",
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1048576,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Muse Spark 1.3 is an update to Meta's Muse Spark reasoning model that improves long-horizon agentic and coding work, follows long instructions more reliably, and adds a max reasoning setting.",
      "highlights": [
        "Available with max reasoning on Muse Code and the Meta Model API",
        "Takes fewer unnecessary turns and is less verbose in coding than Muse Spark 1.2",
        "Meta reports stronger adversarial robustness, including against prompt injection",
        "API docs list a 1,048,576-token context window; audio input is not fully supported"
      ],
      "source_title": "Introducing Muse Spark 1.3",
      "source_url": "https://research.meta.ai/blog/introducing-muse-spark-1-3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#muse-spark-1-3"
    },
    {
      "slug": "claude-fable-5-1",
      "date": "2026-09-01",
      "model": "Claude Fable 5.1",
      "lab": "Anthropic",
      "lab_slug": "anthropic",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Cloud platforms"
      ],
      "summary": "Claude Fable 5.1 is the successor to Claude Fable 5 for long-running agentic coding, knowledge work, and research. It is the generally available version of the model behind Claude Mythos 5.1, with safeguards for cybersecurity and biology.",
      "highlights": [
        "API model ID claude-fable-5-1; 1M-token context window and 128K max output tokens",
        "Claude Mythos 5.1, the same model with fewer safeguards, is limited to vetted organizations",
        "tool_choice types any and tool are not supported",
        "Available on all platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure"
      ],
      "source_title": "Introducing Claude Fable 5.1 and Claude Mythos 5.1",
      "source_url": "https://www.anthropic.com/claude-fable-and-mythos-5-1",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#claude-fable-5-1"
    },
    {
      "slug": "muse-voice-transcribe",
      "date": "2026-09-01",
      "model": "Muse Voice Transcribe",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Muse Voice Transcribe is Meta Superintelligence Labs' first real-time audio perception model, combining streaming speech recognition, speaker diarization and endpointing in one model.",
      "highlights": [
        "Diarization for more than 20 speakers",
        "Trained on 70+ languages, with 25 extensively verified, and handles code-switching",
        "Available in the Meta Model API, Meta AI for Mac and Muse Code",
        "API model ID muse-voice-transcribe-1.0"
      ],
      "source_title": "Introducing Muse Voice Transcribe",
      "source_url": "https://research.meta.ai/blog/introducing-muse-voice-transcribe",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#muse-voice-transcribe"
    },
    {
      "slug": "pii-tracer",
      "date": "2026-09-01",
      "model": "PII-Tracer",
      "lab": "Perplexity",
      "lab_slug": "perplexity",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "0.6B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "A compact 0.6B model for detecting personally identifiable information in long, multilingual conversations, released with the PII-TRACE benchmark and built for the local privacy gate in Perplexity Computer's hybrid compute on Mac.",
      "highlights": [
        "Bidirectional Qwen3 encoder initialized from pplx-embed-v1-0.6b, with token-tagging and sensitivity heads",
        "Tags 9 PII categories, including names, account numbers, addresses, emails and secrets",
        "Released with PII-TRACE, a 13-language benchmark for PII detection across conversations",
        "Weights on Hugging Face under the MIT license, plus GGUF and MLX builds"
      ],
      "source_title": "PII-TRACE: Detecting Personal Data Before It Leaves a Device",
      "source_url": "https://www.perplexity.ai/hub/blog/pii-trace-detecting-personal-data-before-it-leaves-the-device",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/09#pii-tracer"
    },
    {
      "slug": "hy4-preview",
      "date": "2026-08-28",
      "model": "Hy4 preview",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": "MoE: 770B total, 49B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "API",
        "Consumer app"
      ],
      "summary": "Hy4 preview is an early open-weights version of Tencent's Hy4 flagship, a mixture-of-experts model aimed at long-horizon software engineering, document-heavy office work, game development, and scientific research.",
      "highlights": [
        "770B total parameters with 49B active and a 1M-token context window",
        "Weights on Hugging Face, ModelScope, GitHub, and AtomGit",
        "Available on Tencent Cloud and OpenRouter and in CodeBuddy, WorkBuddy, Yuanbao, and ima"
      ],
      "source_title": "Introducing Hy4 preview",
      "source_url": "https://hy.tencent.com/research/hy4-preview",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#hy4-preview"
    },
    {
      "slug": "cohere-parse",
      "date": "2026-08-27",
      "model": "Cohere Parse",
      "lab": "Cohere",
      "lab_slug": "cohere",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 8000,
      "parameters": "2.3B",
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms",
        "Enterprise"
      ],
      "summary": "Document parsing model that turns complex documents into structured Markdown, extracting text in reading order along with tables, forms, images, captions and layout locations.",
      "highlights": [
        "Model ID parse-v5.0",
        "Outputs HTML tables, bounding boxes and image descriptions",
        "Available on Microsoft Foundry, AWS SageMaker and Model Vault"
      ],
      "source_title": "Meet Cohere Parse",
      "source_url": "https://docs.cohere.com/v2/changelog/parse",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#cohere-parse"
    },
    {
      "slug": "gemini-omni-1-1-flash",
      "date": "2026-08-27",
      "model": "Gemini Omni 1.1 Flash",
      "lab": "Google",
      "lab_slug": "google",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "video",
        "audio",
        "text"
      ],
      "context_window": 131072,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms"
      ],
      "summary": "Gemini Omni 1.1 Flash is the generally available, production-ready update to Gemini Omni Flash, adding video extension, first and last frame interpolation, and output resolution control up to 4K.",
      "highlights": [
        "Model ID gemini-omni-1.1-flash in the Gemini API and Google AI Studio",
        "Extends scenes using up to 10 seconds of prior context",
        "Adds 360p previews alongside 720p, 1080p and 4K output, with 1080p and 4K produced by upscaling"
      ],
      "source_title": "Build with Gemini Omni 1.1 Flash",
      "source_url": "https://blog.google/innovation-and-ai/technology/developers-tools/build-with-gemini-omni-1-1-flash/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#gemini-omni-1-1-flash"
    },
    {
      "slug": "midjourney-v8-2-edit",
      "date": "2026-08-27",
      "model": "Midjourney V8.2 Edit Model",
      "lab": "Midjourney",
      "lab_slug": "midjourney",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "First V8.2 image edit model, supporting instruction-based editing, generation from up to four reference images, inpainting and outpainting, with personalization, moodboards and style references.",
      "highlights": [
        "Replaces Omni Reference, Character Reference and Retexture",
        "Available on midjourney.com, alpha.midjourney.com and Discord via --edit",
        "An image quality update to the edit model followed days later"
      ],
      "source_title": "Edit Model for V8",
      "source_url": "https://updates.midjourney.com/edit-model-for-v8/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#midjourney-v8-2-edit"
    },
    {
      "slug": "gemini-3-5-transcribe",
      "date": "2026-08-26",
      "model": "Gemini 3.5 Transcribe",
      "lab": "Google",
      "lab_slug": "google",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Gemini 3.5 Transcribe is a speech-to-text model built on Gemini's audio understanding, released with a streaming Live API variant for voice agents, live captioning and post-call analytics.",
      "highlights": [
        "Two models: gemini-3.5-transcribe (non-streaming) and gemini-3.5-transcribe-live (bidirectional streaming)",
        "Speaker diarization, word-level timestamps and custom vocabulary biasing of up to 1,000 terms",
        "Google describes it as a major advancement over its previous transcription model, Chirp 3"
      ],
      "source_title": "Introducing Gemini 3.5 Transcribe",
      "source_url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-5-transcribe/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#gemini-3-5-transcribe"
    },
    {
      "slug": "glm-5-3-flash",
      "date": "2026-08-26",
      "model": "GLM-5.3-Flash",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video",
        "file"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": "MoE: 320B total, 18B active",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "GLM-5.3-Flash is the first natively multimodal model in Z.ai's GLM-5 series, trained from a new base model. It is the first GLM model to combine sparse and linear attention in a hybrid architecture that cuts long-context serving costs.",
      "highlights": [
        "Native visual capabilities let it observe interfaces, rendered output, and interaction feedback",
        "Up to 128K output tokens",
        "Extends beyond coding to office document and financial research workflows",
        "Weights on Hugging Face and ModelScope under the MIT license"
      ],
      "source_title": "GLM-5.3-Flash: Frontier Intelligence, Flash Cost",
      "source_url": "https://z.ai/blog/glm-5.3-flash",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#glm-5-3-flash"
    },
    {
      "slug": "qwen3-8-flash-next",
      "date": "2026-08-26",
      "model": "Qwen3.8-Flash-Next",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 262144,
      "parameters": "MoE: 125B total, 6B active, plus 51B n-gram embedding",
      "model_license": "Qwen Community License 1.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Qwen3.8-Flash-Next is an open-weights multimodal mixture-of-experts model that previews the architecture planned for Qwen4. It combines Gated DeltaNet with Qwen Sparse Attention and adds Gated Residual and N-gram Embedding.",
      "highlights": [
        "Native 262,144-token context, extensible to 1,000,000 tokens with YaRN",
        "Qwen says training cost about 1/9 of Qwen3.7-Plus",
        "Weights on Hugging Face and ModelScope under the Qwen Community License 1.0"
      ],
      "source_title": "Qwen3.8-Flash-Next: A New Architecture, Towards Ultimate Cost-Efficiency",
      "source_url": "https://qwen.ai/blog?id=qwen3.8-flash-next",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#qwen3-8-flash-next"
    },
    {
      "slug": "granite-4-2",
      "date": "2026-08-25",
      "model": "Granite 4.2",
      "lab": "IBM",
      "lab_slug": "ibm",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 512000,
      "parameters": "3B, 8B and 30B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Granite generation that introduces native chain-of-thought reasoning, with full thinking, non-thinking and low-effort modes in 3B, 8B and 30B models.",
      "highlights": [
        "Supports 128K tokens natively, extendable to 512K",
        "Targets math, coding and agentic tool calling",
        "Released under Apache 2.0"
      ],
      "source_title": "ibm-granite/granite-4.2-8b · Hugging Face",
      "source_url": "https://huggingface.co/ibm-granite/granite-4.2-8b",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#granite-4-2"
    },
    {
      "slug": "deepseek-v4-flash-vision-exp",
      "date": "2026-08-21",
      "model": "DeepSeek-V4-Flash-Vision-Exp",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "DeepSeek-V4-Flash-Vision-Exp is DeepSeek's first experimental multimodal model in the V4 family, adding visual modules to the V4-Flash architecture. DeepSeek says it matches V4-Flash on text tasks while improving multimodal agent performance.",
      "highlights": [
        "API model name deepseek-v4-flash-vision-exp, supporting mixed text and image input",
        "Launched alongside a new Files API for reusing uploaded images",
        "Retired on the API on September 10, 2026, when V4.1-Flash replaced it",
        "Weights published on Hugging Face under the MIT License"
      ],
      "source_title": "DeepSeek-V4-Flash-Vision-Exp Release: Multimodal API Now Live",
      "source_url": "https://api-docs.deepseek.com/news/news260821",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#deepseek-v4-flash-vision-exp"
    },
    {
      "slug": "runway-ruby",
      "date": "2026-08-20",
      "model": "Ruby",
      "lab": "Runway",
      "lab_slug": "runway",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "video"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Native SDR-to-HDR color grading model that converts SDR video from any source into true HDR without re-rendering pixels, delivering HDR10, HLG, ProRes or OpenEXR masters.",
      "highlights": [
        "Accepts SDR clips up to 30 seconds and up to 4K",
        "Keeps source pixels and audio intact",
        "Available in Tool Mode, Workflows and Runway Dev for Max and higher plans"
      ],
      "source_title": "Product Updates & Changelog | Runway AI",
      "source_url": "https://runway.com/changelog",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#runway-ruby"
    },
    {
      "slug": "glm-5-3",
      "date": "2026-08-14",
      "model": "GLM-5.3",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": "MoE: 744B total, 40B active",
      "model_license": "GLM-5.3 License",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "GLM-5.3 is Z.ai's flagship coding model, which uses the same base model as GLM-5.2 with all gains coming from post-training. Z.ai reports emergent cybersecurity capability in vulnerability discovery and exploitation.",
      "highlights": [
        "Text-only input with up to 128K output tokens",
        "Reasoning is always enabled, with low, high, and max effort levels",
        "Listed in Z.ai API release notes on August 18, 2026; weights released after safety evaluation",
        "Weights under the GLM-5.3 License, which adds a security review for very large model-as-a-service providers"
      ],
      "source_title": "GLM-5.3: Frontier Coding with Emergent Cyber Capabilities",
      "source_url": "https://z.ai/blog/glm-5.3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#glm-5-3"
    },
    {
      "slug": "pika-music",
      "date": "2026-08-14",
      "model": "Pika Music",
      "lab": "Pika",
      "lab_slug": "pika",
      "type": "music",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Music generation model that turns text, lyrics, a reference voice, an existing track or combinations of these into finished songs through a single workflow.",
      "highlights": [
        "Announced August 14, 2026 with the Pika Audio model family",
        "Supports voice conditioning and music references",
        "Available through the Pika API Club"
      ],
      "source_title": "Introducing Pika Music",
      "source_url": "https://pika.art/blog/pika-music",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#pika-music"
    },
    {
      "slug": "pika-sfx",
      "date": "2026-08-14",
      "model": "Pika SFX",
      "lab": "Pika",
      "lab_slug": "pika",
      "type": "music",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Text-to-sound-effects model that generates focused, ready-to-use effects in real time, from single events to timed sequences, while avoiding unrequested speech, music and noise.",
      "highlights": [
        "Announced August 14, 2026 with the Pika Audio model family",
        "Handles material, space and perspective cues",
        "Available through the Pika API Club"
      ],
      "source_title": "Now Hear This: Pika SFX",
      "source_url": "https://pika.art/blog/pika-sfx",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#pika-sfx"
    },
    {
      "slug": "pika-soundtrack",
      "date": "2026-08-14",
      "model": "Pika Soundtrack",
      "lab": "Pika",
      "lab_slug": "pika",
      "type": "music",
      "weights": "proprietary",
      "input_modalities": [
        "video",
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Video-to-audio model in Pika's first audio family that turns video into a synchronized full-scene soundscape of motion-aware sound effects, voice, music and ambience.",
      "highlights": [
        "Announced August 14, 2026 with the Pika Audio model family",
        "Optional instructions steer what to emphasize or leave out",
        "Available through the Pika API Club"
      ],
      "source_title": "Introducing Pika Soundtrack",
      "source_url": "https://pika.art/blog/pika-soundtrack",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#pika-soundtrack"
    },
    {
      "slug": "pika-speech",
      "date": "2026-08-14",
      "model": "Pika Speech",
      "lab": "Pika",
      "lab_slug": "pika",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": "3B",
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Text-to-speech model built as a 3B flow-matching transformer that generates 48 kHz speech in a few denoising steps and clones voices from a few seconds of reference audio.",
      "highlights": [
        "3B-parameter flow-matching transformer",
        "Up to five minutes of speech per request",
        "Controllable style, pace and duration"
      ],
      "source_title": "Say Hello to Pika Speech",
      "source_url": "https://pika.art/blog/pika-speech",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#pika-speech"
    },
    {
      "slug": "qwen3-8-27b",
      "date": "2026-08-14",
      "model": "Qwen3.8-27B",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 262144,
      "parameters": "27B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Qwen3.8-27B is a dense 27B open-weights model in the Qwen3.8 series, released on Hugging Face and ModelScope after Qwen3.8-Max. It is a causal language model with a vision encoder.",
      "highlights": [
        "Native 262,144-token context, extensible to 1,000,000 tokens",
        "Released under the Apache 2.0 license",
        "Follows the Qwen3.8-2.4T-A95B open-weights release on August 12, 2026"
      ],
      "source_title": "Qwen3.8 (GitHub)",
      "source_url": "https://github.com/QwenLM/Qwen3.8",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#qwen3-8-27b"
    },
    {
      "slug": "deepseek-v4-pro-0813",
      "date": "2026-08-13",
      "model": "DeepSeek-V4-Pro (GA, 0813)",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": "MoE: 1.6T total, 49B active",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "DeepSeek-V4-Pro-0813 is the general availability release of DeepSeek-V4-Pro, superseding the April preview with greatly enhanced agentic capabilities. It keeps the preview's model structure and adds an attached DSpark speculative decoding module.",
      "highlights": [
        "Rolled out on the DeepSeek app, web, and API under the unchanged deepseek-v4-pro model name",
        "Added low, high, and max reasoning effort levels for V4-Pro and V4-Flash",
        "DeepSeek API gained native OpenAI Responses API support with this release"
      ],
      "source_title": "DeepSeek-V4-Pro GA Release",
      "source_url": "https://api-docs.deepseek.com/news/news260813",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#deepseek-v4-pro-0813"
    },
    {
      "slug": "gemini-3-7-flash",
      "date": "2026-08-13",
      "model": "Gemini 3.7 Flash",
      "lab": "Google",
      "lab_slug": "google",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1048576,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Gemini 3.7 Flash is a Gemini workhorse model for coding and agents, released three weeks after 3.6 Flash with substantial improvements across software engineering, knowledge work and web development.",
      "highlights": [
        "Generally available in the Gemini API via Google AI Studio and Android Studio, and in Google Antigravity",
        "Powers Gemini Spark, the personal agent in the Gemini app for Google AI Pro and Ultra subscribers",
        "Replaced by Gemini 3.8 Flash, with requests routed to 3.8 Flash from Oct 8, 2026"
      ],
      "source_title": "Gemini 3.7 Flash: our most intelligent workhorse model",
      "source_url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-3-7-flash/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#gemini-3-7-flash"
    },
    {
      "slug": "grok-4-6",
      "date": "2026-08-12",
      "model": "Grok 4.6",
      "lab": "xAI",
      "lab_slug": "xai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 500000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Grok 4.6 builds on Grok 4.5 with a focus on long-running agents and more ambitious interactive and visual work, such as turning a broad product idea into a working first version of an application.",
      "highlights": [
        "API model ID: grok-4.6, with a 500k-token context window",
        "Available at launch in Cursor, Grok Build, the API, OpenRouter, Vercel, and Cloudflare",
        "Reasoning effort settings: low, medium, high (default), and xhigh",
        "A fast variant is also offered"
      ],
      "source_title": "Introducing Grok 4.6 | SpaceXSI",
      "source_url": "https://x.ai/news/grok-4-6",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#grok-4-6"
    },
    {
      "slug": "lfm2-5-vl-3b",
      "date": "2026-08-12",
      "model": "LFM2.5-VL-3B",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32768,
      "parameters": "3.1B",
      "model_license": "LFM Open License v1.0",
      "available_at_launch": [
        "Consumer app",
        "Open weights"
      ],
      "summary": "An updated 3B vision-language model building on LFM2-VL-3B with improved screen and UI understanding, grounding, function calling for text and image inputs, and multi-image input, answering directly without a reasoning phase.",
      "highlights": [
        "Built on the LFM2.5-2.6B-Base backbone with a 128,000-token vocabulary",
        "Vision pretraining scaled by 4x tokens",
        "32,768-token context length per model card",
        "Available on Hugging Face and Liquid Playground"
      ],
      "source_title": "LFM2.5-VL-3B: A Better and Faster Vision-Language Model for the Edge",
      "source_url": "https://www.liquid.ai/blog/lfm2-5-vl-3b",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#lfm2-5-vl-3b"
    },
    {
      "slug": "north-micro-vision",
      "date": "2026-08-12",
      "model": "North Micro Vision Instruct",
      "lab": "Cohere",
      "lab_slug": "cohere",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "2.4B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Compact open-weight vision-language model from Cohere with native-resolution image support, intended as a foundation for prototyping, task-specific fine-tuning and specialized multimodal applications.",
      "highlights": [
        "About 2B language model plus a 400M vision encoder",
        "Validated for multimodal prompts up to 8K tokens",
        "Fine-tuning recipes available from NVIDIA and Axolotl"
      ],
      "source_title": "Meet North Micro Vision: A 2.4B Native-Resolution Vision-Language Model",
      "source_url": "https://huggingface.co/blog/CohereLabs/meet-north-micro-vision-instruct",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#north-micro-vision"
    },
    {
      "slug": "ltx-2-5",
      "date": "2026-08-11",
      "model": "LTX-2.5",
      "lab": "Lightricks",
      "lab_slug": "lightricks",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video",
        "audio"
      ],
      "output_modalities": [
        "video",
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "LTX-2.x Community License",
      "available_at_launch": [
        "Open weights",
        "API",
        "Consumer app"
      ],
      "summary": "LTX model superseding LTX-Video and LTX-2, adding native multi-shot generation, a diffusion video decoder, a custom Gemma 4 12B text encoder and a raw pretrained checkpoint for fine-tuning.",
      "highlights": [
        "Native multi-shot scenes with consistent characters and voices",
        "Native EXR workflow in ACES and DaVinci Wide Gamut",
        "Released as open weights and in the LTX API"
      ],
      "source_title": "Release Notes | LTX",
      "source_url": "https://ltx.io/release-notes",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#ltx-2-5"
    },
    {
      "slug": "mai-code-1-1-flash",
      "date": "2026-08-11",
      "model": "MAI-Code-1.1-Flash",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "coding",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "MAI-Code-1.1-Flash is Microsoft AI's updated small coding model for GitHub Copilot, focused on CLI and .NET tasks and using fewer tokens per task than MAI-Code-1-Flash. A quantized on-device version followed in October 2026.",
      "highlights": [
        "In production in GitHub Copilot at launch",
        "October 7, 2026 update added a 3-bit quantized version that runs locally with a full 256K context window",
        "Local use recommended on devices with more than 120GB of RAM"
      ],
      "source_title": "MAI-Code-1.1-Flash: Better, faster, at a quarter of the cost",
      "source_url": "https://microsoft.ai/news/mai-code-1-1-flash-br-better-faster-at-a-quarter-of-the-cost/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#mai-code-1-1-flash"
    },
    {
      "slug": "nemotron-3-5-lightning",
      "date": "2026-08-11",
      "model": "Nemotron 3.5 Lightning",
      "lab": "NVIDIA",
      "lab_slug": "nvidia",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": "MoE: 30B total, 3B active",
      "model_license": "OpenMDW-1.1",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Nemotron 3.5 language model built on a hybrid Mamba-2 and mixture-of-experts architecture, released alongside several speculative decoding methods to speed up text generation.",
      "highlights": [
        "Up to 1M tokens of context, with 256K used for single H100 deployment",
        "Trained between December 2025 and May 2026",
        "Released under the OpenMDW 1.1 license"
      ],
      "source_title": "nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16 · Hugging Face",
      "source_url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#nemotron-3-5-lightning"
    },
    {
      "slug": "mai-image-2-6",
      "date": "2026-08-10",
      "model": "MAI-Image-2.6",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "MAI-Image-2.6 is Microsoft AI's text-to-image and image editing model succeeding MAI-Image-2.5, with stronger text rendering, portraits, 3D imagery and commercial photorealistic output. Developer access in Microsoft Foundry followed on September 4, 2026.",
      "highlights": [
        "Brought to developers in Microsoft Foundry on September 4, 2026",
        "Generally available in Foundry per an October 27, 2026 update",
        "Supports multiple image references, web grounding and control over format and resolution"
      ],
      "source_title": "MAI Image 2.6 launches at No. 2 on Arena ahead of Google, Meta and xAI",
      "source_url": "https://microsoft.ai/news/mai-image-2-6-launches-at-no-2-on-arena-ahead-of-google-meta-and-xai/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#mai-image-2-6"
    },
    {
      "slug": "muse-glimmer",
      "date": "2026-08-10",
      "model": "Muse Glimmer",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 131072,
      "parameters": "30B (dense)",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Muse Glimmer is a 30B open-weights model from Meta Superintelligence Labs, distilled from Muse Spark and built for always-on local agent workflows on consumer hardware.",
      "highlights": [
        "Released under Apache 2.0, the most permissive license Meta has used for an open model",
        "Accepts interleaved text and images through a dedicated perception encoder",
        "Trained on Muse Spark outputs using logit distillation",
        "Weights on Hugging Face, including GGUF and ExecuTorch builds"
      ],
      "source_title": "Introducing Muse Glimmer: An Open Agentic Model That Runs on Your Device",
      "source_url": "https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#muse-glimmer"
    },
    {
      "slug": "grok-imagine-image-2-0",
      "date": "2026-08-07",
      "model": "Grok Imagine Image 2.0",
      "lab": "xAI",
      "lab_slug": "xai",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Imagine Image 2.0 is SpaceXSI's image generation and editing model, built for close instruction following, designer-style typography and layout, and preserving inputs across generations and edits. It became the Quality Mode on grok.com/imagine.",
      "highlights": [
        "API model ID: grok-imagine-image-2.0",
        "Generally available as Quality Mode on grok.com/imagine and the iOS and Android apps",
        "Multi-reference editing accepts up to 5 input images per generation",
        "Adds region editing, segmentation, background removal, and smart resize"
      ],
      "source_title": "Imagine Image 2.0 | SpaceXSI",
      "source_url": "https://x.ai/news/grok-imagine-image-2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#grok-imagine-image-2-0"
    },
    {
      "slug": "wan-animate-2",
      "date": "2026-08-07",
      "model": "Wan-Animate-2",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": "14B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Wan-Animate-2 is an end-to-end character animation model that directly consumes driving videos in a redesigned Diffusion Transformer. It targets high-fidelity motion and strong identity preservation.",
      "highlights": [
        "Released as Base and Distillation weights on August 7, 2026",
        "Published on Hugging Face as Wan2.2-Animate-2-14B",
        "Released under the Apache 2.0 license"
      ],
      "source_title": "Wan-Animate-2 (GitHub)",
      "source_url": "https://github.com/Wan-Video/Wan-Animate-2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#wan-animate-2"
    },
    {
      "slug": "muse-spark-1-2",
      "date": "2026-08-05",
      "model": "Muse Spark 1.2",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "coding",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video",
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1048576,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Muse Spark 1.2 is a coding-focused update to Muse Spark 1.1, launched alongside Muse Code, Meta's terminal coding agent, with gains in code generation, debugging and codebase understanding.",
      "highlights": [
        "Launched with Muse Code, a beta terminal coding agent",
        "Available in Muse Code and the Meta Model API with expanded global access",
        "Meta scaled up training compute on coding tasks",
        "API docs list a 1,048,576-token context window"
      ],
      "source_title": "Introducing Muse Code and Muse Spark 1.2",
      "source_url": "https://research.meta.ai/blog/introducing-muse-code-and-muse-spark-1-2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#muse-spark-1-2"
    },
    {
      "slug": "seedrealtime",
      "date": "2026-08-05",
      "model": "SeedRealtime",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "audio",
        "video"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "SeedRealtime is a native audio-visual full-duplex model that jointly understands audio, video, and timing in continuous streams, letting the Doubao app hold real-time video calls in which it watches, listens, and speaks, sometimes proactively.",
      "highlights": [
        "Unified architecture natively fuses audio, video, and text",
        "Decides when to speak without relying on an external voice activity detector",
        "Can weave tool calls into its spoken responses",
        "Fully rolled out in the Doubao app's video call mode"
      ],
      "source_title": "SeedRealtime Audio-Visual Full-Duplex LLM Released: Toward Omni-Modal Natural Interaction",
      "source_url": "https://seed.bytedance.com/en/blog/seedrealtime-audio-visual-full-duplex-llm-released-toward-omni-modal-natural-interaction",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#seedrealtime"
    },
    {
      "slug": "lfm2-5-2-6b",
      "date": "2026-08-04",
      "model": "LFM2.5-2.6B",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 131072,
      "parameters": "2.6B",
      "model_license": "LFM Open License v1.0",
      "available_at_launch": [
        "Consumer app",
        "Open weights"
      ],
      "summary": "An agentic 2.6B model that runs on-device, pre-trained on about 34T tokens with a 128K context extension, a vocabulary doubled to 128K, and a four-stage post-training pipeline ending in agentic reinforcement learning.",
      "highlights": [
        "131,072-token context length per model card",
        "Post-training: SFT, teacher specialization, multi-domain on-policy distillation, agentic RL",
        "Base (LFM2.5-2.6B-Base) and post-trained checkpoints on Hugging Face"
      ],
      "source_title": "LFM2.5-2.6B: Deploy Agents Everywhere",
      "source_url": "https://www.liquid.ai/blog/lfm2-5-2-6b",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#lfm2-5-2-6b"
    },
    {
      "slug": "shieldstral",
      "date": "2026-08-04",
      "model": "Shieldstral 1.0",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32000,
      "parameters": "3.8B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Compact multimodal safety classifier that takes a plain-language policy question at inference time and returns a calibrated safety score for text and images without retraining.",
      "highlights": [
        "Runs on a single 16GB NVIDIA GPU",
        "Released as an inaugural member of the Open Secure AI Alliance with NVIDIA",
        "In public preview on Mistral's platform"
      ],
      "source_title": "Introducing Shieldstral. | Mistral",
      "source_url": "https://mistral.ai/news/shieldstral",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#shieldstral"
    },
    {
      "slug": "qwen3-8-max",
      "date": "2026-08-03",
      "model": "Qwen3.8-Max",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 262144,
      "parameters": "MoE: 2.4T total, 95B active",
      "model_license": "Qwen3.8-Max License",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "Qwen3.8-Max is the most capable model in Alibaba's Qwen family to date, built on the Qwen3.5 architecture for coding, work, research, and long-horizon tasks. It is the first Qwen-Max-class model with open weights.",
      "highlights": [
        "Weights published as Qwen3.8-2.4T-A95B on Hugging Face and ModelScope on August 12, 2026",
        "Native 262,144-token context, extensible to about 1M tokens",
        "API available on QwenCloud at launch"
      ],
      "source_title": "Qwen3.8-Max: A New Bar for Coding and Cowork",
      "source_url": "https://qwen.ai/blog?id=qwen3.8",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/08#qwen3-8-max"
    },
    {
      "slug": "deepseek-v4-flash-0731",
      "date": "2026-07-31",
      "model": "DeepSeek-V4-Flash (0731)",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": "MoE: 284B total, 13B active",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "DeepSeek-V4-Flash-0731 is the official release of DeepSeek-V4-Flash, superseding the April preview with substantially enhanced agentic capabilities. It keeps the preview's architecture and size and was re-post-trained, with a speculative decoding module attached.",
      "highlights": [
        "Released to the deepseek-v4-flash API as a public beta; the app, web, and V4-Pro API were unchanged",
        "Supports low, high, and max reasoning effort levels",
        "Weights on Hugging Face as DeepSeek-V4-Flash-0731 under the MIT License"
      ],
      "source_title": "Change Log | DeepSeek API Docs",
      "source_url": "https://api-docs.deepseek.com/updates",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#deepseek-v4-flash-0731"
    },
    {
      "slug": "minimax-h3",
      "date": "2026-07-31",
      "model": "MiniMax H3",
      "lab": "MiniMax",
      "lab_slug": "minimax",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video",
        "audio"
      ],
      "output_modalities": [
        "video",
        "audio"
      ],
      "context_window": null,
      "parameters": "33B dense Omni Transformer",
      "model_license": "MiniMax H3 Community License Agreement",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "MiniMax H3 is a general-purpose multimodal generation model that reads combined text, image, video, and audio context and generates video with native stereo sound, up to 15 seconds long at up to 2K resolution.",
      "highlights": [
        "Weights open-sourced on August 3, 2026 as first-and-last-frame and omni-reference checkpoints",
        "Core H3-Omni-Transformer is a 33B-parameter dense single-stream Transformer",
        "Outputs 4 to 15 second clips at 24 fps with 32 kHz stereo audio",
        "API variants include H3 Max, H3 Highspeed, and H3 Lite"
      ],
      "source_title": "MiniMax H3: An Open Model Breaking the Boundaries Between Tasks and Modalities",
      "source_url": "https://www.minimax.io/blog/minimax-h3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#minimax-h3"
    },
    {
      "slug": "seedance-2-5",
      "date": "2026-07-31",
      "model": "Seedance 2.5",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video",
        "audio"
      ],
      "output_modalities": [
        "video",
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Seedance 2.5 is ByteDance Seed's video creation model built on the Seedance 2.0 joint audio-video architecture, extending single-pass generation to 30 seconds with multi-round extension and upgraded multimodal referencing and editing.",
      "highlights": [
        "Generates up to 30 seconds of audio and video per pass, up from 15 seconds",
        "Accepts up to 30 images, 10 video clips, and 10 audio clips as references",
        "Rolling out on Jimeng AI and Doubao; BytePlus offers it as Dreamina Seedance 2.5",
        "Supports timestamp-based editing, per Volcano Engine Ark release notes"
      ],
      "source_title": "One-take Creation, Flexible Referencing: Introducing Seedance 2.5",
      "source_url": "https://seed.bytedance.com/en/blog/one-take-creation-flexible-referencing-introducing-seedance-2-5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#seedance-2-5"
    },
    {
      "slug": "gemini-robotics-er-2",
      "date": "2026-07-30",
      "model": "Gemini Robotics ER 2",
      "lab": "Google",
      "lab_slug": "google",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 131072,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms"
      ],
      "summary": "Gemini Robotics ER 2 is Google DeepMind's most capable embodied reasoning model, acting as a high-level brain that lets robots understand the physical world, plan multi-step tasks, track progress from video and coordinate with other robots.",
      "highlights": [
        "Publicly available in the Gemini API and Google AI Studio, and in private preview on Gemini Enterprise Agent Platform",
        "Ships as gemini-robotics-er-2-preview plus a streaming variant for the Live API",
        "Supports function calling with blocking behavior for physical robot actions"
      ],
      "source_title": "Gemini Robotics ER 2",
      "source_url": "https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-robotics-er-2/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#gemini-robotics-er-2"
    },
    {
      "slug": "inkling-small",
      "date": "2026-07-30",
      "model": "Inkling-Small",
      "lab": "Thinking Machines Lab",
      "lab_slug": "thinking-machines-lab",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": "MoE: 276B total, 12B active",
      "model_license": "Apache 2.0",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "A smaller open-weights sibling of Inkling that Thinking Machines Lab says reaches comparable performance at a quarter of the size, with native reasoning over text, images and audio.",
      "highlights": [
        "276B total parameters, 12B active; 42 layers routing each token to 6 of 256 experts plus 2 shared experts",
        "Context window up to 1M tokens",
        "Ahead of Inkling on reasoning and agentic coding in the lab's results; Inkling kept the edge on knowledge and factuality",
        "Weights on Hugging Face (original and NVFP4) under Apache 2.0; fine-tuning on Tinker and chat in the Tinker Playground"
      ],
      "source_title": "Introducing Inkling-Small",
      "source_url": "https://thinkingmachines.ai/news/inkling-small/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#inkling-small"
    },
    {
      "slug": "grok-voice-2-0",
      "date": "2026-07-29",
      "model": "Grok Voice 2.0",
      "lab": "xAI",
      "lab_slug": "xai",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio",
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Grok Voice 2.0 is SpaceXSI's next-generation speech-to-speech model, improving speech reasoning, transcription accuracy, conversational dynamics, and tool use over Grok Voice 1.0 while using reasoning tokens more efficiently.",
      "highlights": [
        "API model ID: grok-voice-2.0, in the Speech to Speech API",
        "Set to become the grok-voice-latest target on August 5, 2026",
        "Reasons through queries while speaking, with no added latency per SpaceXSI",
        "Deprecated on October 14, 2026 in favor of Grok Voice 3.0"
      ],
      "source_title": "Introducing Grok Voice 2.0 | SpaceXSI",
      "source_url": "https://x.ai/news/grok-voice-think-fast-2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#grok-voice-2-0"
    },
    {
      "slug": "lyria-3-5",
      "date": "2026-07-29",
      "model": "Lyria 3.5",
      "lab": "Google",
      "lab_slug": "google",
      "type": "music",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Lyria 3.5 is a Google DeepMind music generation model that launched first in Google Flow Music. It improves musicality, lyrics and vocal quality over Lyria 3 and adds easier control over tempo and track duration.",
      "highlights": [
        "Rolled out first in Google Flow Music on July 29, 2026",
        "More natural melodies, stronger lyric adherence and more expressive vocals",
        "Generally available in the Gemini API as lyria-3.5 from September 3, 2026",
        "Available in the Gemini app globally from September 4, 2026"
      ],
      "source_title": "We're launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control.",
      "source_url": "https://blog.google/innovation-and-ai/models-and-research/google-labs/lyria-3-5/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#lyria-3-5"
    },
    {
      "slug": "gpt-transcribe",
      "date": "2026-07-28",
      "model": "GPT Transcribe and GPT Live Transcribe",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio",
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "GPT Transcribe and GPT Live Transcribe are speech-to-text models for the API: GPT Transcribe handles audio files and committed Realtime turns, while GPT Live Transcribe provides low-latency streaming transcription with tunable latency.",
      "highlights": [
        "API model IDs gpt-transcribe (v1/audio/transcriptions) and gpt-live-transcribe (v1/realtime)",
        "Both support free-form transcription context, keyword hints, and multiple expected input languages",
        "Named as migration targets for whisper-1 and the gpt-4o transcribe models, which shut down February 26, 2027"
      ],
      "source_title": "Changelog | OpenAI API",
      "source_url": "https://developers.openai.com/api/docs/changelog",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#gpt-transcribe"
    },
    {
      "slug": "lfm2-5-encoders",
      "date": "2026-07-28",
      "model": "LFM2.5-Encoder-230M and LFM2.5-Encoder-350M",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "embedding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "embedding"
      ],
      "context_window": 8192,
      "parameters": "230M; 350M",
      "model_license": "LFM Open License v1.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Two bidirectional masked-language encoders on the LFM2 hybrid architecture, initialized from LFM2.5-230M and LFM2.5-350M and meant to be fine-tuned for classification, natural language understanding, token-level tasks, and retrieval.",
      "highlights": [
        "Native 8,192-token context length",
        "Trained with a masked language modeling objective that masks 30% of tokens",
        "Demo fine-tunes include PII detection, spell checking, and a policy linter",
        "Available on Hugging Face"
      ],
      "source_title": "LFM2.5-Encoders: Fast at Long Context, Even on CPU",
      "source_url": "https://www.liquid.ai/blog/lfm2-5-encoders",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#lfm2-5-encoders"
    },
    {
      "slug": "claude-opus-5",
      "date": "2026-07-24",
      "model": "Claude Opus 5",
      "lab": "Anthropic",
      "lab_slug": "anthropic",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Cloud platforms"
      ],
      "summary": "Claude Opus 5 succeeds Claude Opus 4.8, and Anthropic says it comes close to the capability of Claude Fable 5. It targets everyday coding, knowledge work, and long-running agent workloads.",
      "highlights": [
        "API model ID claude-opus-5; 1M-token context window and 128K max output tokens",
        "New default model on Claude Max",
        "Fast mode runs at around 2.5 times the default speed",
        "Optional automatic fallbacks route safety-flagged API requests to another model instead of blocking them"
      ],
      "source_title": "Introducing Claude Opus 5",
      "source_url": "https://www.anthropic.com/news/claude-opus-5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#claude-opus-5"
    },
    {
      "slug": "midjourney-v8-2",
      "date": "2026-07-24",
      "model": "Midjourney V8.2",
      "lab": "Midjourney",
      "lab_slug": "midjourney",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "V8 update focused on aesthetics, image quality and personalization, producing bolder and more sophisticated images and reducing low-quality outliers, released as the default model.",
      "highlights": [
        "Released as the default version on July 24, 2026",
        "Larger image pool for building personalization profiles",
        "Followed on Aug 27, 2026 by an Edit Model that replaced Omni Reference"
      ],
      "source_title": "Version - Midjourney",
      "source_url": "https://docs.midjourney.com/hc/en-us/articles/32199405667853-Version",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#midjourney-v8-2"
    },
    {
      "slug": "flux-3-video",
      "date": "2026-07-23",
      "model": "FLUX 3 (Video)",
      "lab": "Black Forest Labs",
      "lab_slug": "black-forest-labs",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video",
        "audio"
      ],
      "output_modalities": [
        "video",
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms"
      ],
      "summary": "First FLUX multimodal foundation model, jointly trained on images, video and audio, announced in early access with video generation featuring synchronized audio, multilingual dialogue and keyframe control.",
      "highlights": [
        "Clips up to 20 seconds with native audio",
        "Video generation became generally available via the BFL API on August 4, 2026",
        "2K and 4K output added to the video endpoint on September 10, 2026"
      ],
      "source_title": "FLUX 3 - Real World Models: Towards Multimodal Flow Models as the Backbone of Visual Intelligence",
      "source_url": "https://bfl.ai/blog/flux-3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#flux-3-video"
    },
    {
      "slug": "mai-image-2-5-pro",
      "date": "2026-07-23",
      "model": "MAI-Image-2.5-Pro",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "MAI-Image-2.5-Pro is Microsoft AI's highest-fidelity image model at launch, released in public preview for hero imagery, detailed editing and precise in-image text rendering, alongside the existing MAI-Image-2.5 and 2.5-Flash models.",
      "highlights": [
        "Released in public preview",
        "Targets hero imagery, detailed editing and precise in-image text rendering",
        "Sits above MAI-Image-2.5 and MAI-Image-2.5-Flash as the highest-fidelity option"
      ],
      "source_title": "Introducing MAI-Image-2.5-Pro and MAI-Voice-2-Flash",
      "source_url": "https://microsoft.ai/news/introducing-mai-image-2-5-pro-and-mai-voice-2-flash/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#mai-image-2-5-pro"
    },
    {
      "slug": "mai-voice-2-flash",
      "date": "2026-07-23",
      "model": "MAI-Voice-2-Flash",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "MAI-Voice-2-Flash is a faster, lower-cost variant of Microsoft AI's MAI-Voice-2 text-to-speech model for high-volume voice experiences. First previewed at Microsoft Build in June 2026, it entered public preview in July 2026.",
      "highlights": [
        "Released in public preview on July 23, 2026",
        "Microsoft says it is 2x faster than MAI-Voice-2",
        "Announced as coming soon on June 2, 2026"
      ],
      "source_title": "Introducing MAI-Image-2.5-Pro and MAI-Voice-2-Flash",
      "source_url": "https://microsoft.ai/news/introducing-mai-image-2-5-pro-and-mai-voice-2-flash/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#mai-voice-2-flash"
    },
    {
      "slug": "gemini-3-5-flash-lite",
      "date": "2026-07-21",
      "model": "Gemini 3.5 Flash-Lite",
      "lab": "Google",
      "lab_slug": "google",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1048576,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms",
        "Consumer app"
      ],
      "summary": "Gemini 3.5 Flash-Lite is the fastest and most cost-effective 3.5-class Gemini model, designed for high-throughput production traffic and as a low-latency subagent option for high-volume automation.",
      "highlights": [
        "Generally available in the Gemini API, Gemini Enterprise Agent Platform and the Gemini app, and rolling out in Google Search",
        "Significant step up over prior Flash-Lite generations in agentic workflows",
        "Input token limit of 1,048,576 per the Gemini API model page"
      ],
      "source_title": "3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber",
      "source_url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-6-flash-3-5-flash-lite-3-5-flash-cyber/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#gemini-3-5-flash-lite"
    },
    {
      "slug": "gemini-3-6-flash",
      "date": "2026-07-21",
      "model": "Gemini 3.6 Flash",
      "lab": "Google",
      "lab_slug": "google",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1048576,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms",
        "Consumer app"
      ],
      "summary": "Gemini 3.6 Flash is a Gemini workhorse model that improves coding, knowledge work and multimodal performance over 3.5 Flash while using fewer output tokens, fewer reasoning steps and fewer tool calls.",
      "highlights": [
        "Generally available in the Gemini API, Google Antigravity, Gemini Enterprise Agent Platform and the Gemini app",
        "Reduces output verbosity in response to developer feedback on 3.5 Flash",
        "Shipped with enhanced Frontier Safety safeguards for CBRN and cyber offense misuse"
      ],
      "source_title": "3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber",
      "source_url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-6-flash-3-5-flash-lite-3-5-flash-cyber/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#gemini-3-6-flash"
    },
    {
      "slug": "qwen-image-3-0",
      "date": "2026-07-21",
      "model": "Qwen-Image-3.0",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Qwen-Image-3.0 is the third-generation foundational image generation model in the Qwen-Image series. Qwen centers it on realism, with richer content, authentic details, and deeper knowledge.",
      "highlights": [
        "Supports up to 4.5K tokens of input for complex layouts such as newspapers and storyboards",
        "Available to try on Qwen Chat",
        "Successor to Qwen-Image-2.0"
      ],
      "source_title": "Qwen-Image-3.0: Rich Content, Authentic Details, Deep Knowledge",
      "source_url": "https://qwen.ai/blog?id=qwen-image-3.0",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#qwen-image-3-0"
    },
    {
      "slug": "seed-audio-1-0",
      "date": "2026-07-20",
      "model": "Seed Audio 1.0",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Seed Audio 1.0 is an audio creation model that generates speech, sound effects, ambience, and other scene-level audio within one unified framework, coordinating character performance, timing, and background sound from a single prompt.",
      "highlights": [
        "Models voice, sound effects, and ambient sound jointly instead of as separate clips",
        "Keeps voice identity consistent across a broad expressive range",
        "Available through BytePlus at launch"
      ],
      "source_title": "From Speech to Audio Creation | Introducing the Seed Audio 1.0 Audio Creation Model",
      "source_url": "https://seed.bytedance.com/en/blog/from-speech-to-audio-creation-introducing-the-seed-audio-1-0-audio-creation-model",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#seed-audio-1-0"
    },
    {
      "slug": "grok-4-5",
      "date": "2026-07-16",
      "model": "Grok 4.5",
      "lab": "xAI",
      "lab_slug": "xai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 500000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Grok 4.5 is SpaceXSI's model for coding, agentic tasks, and knowledge work, trained alongside Cursor on datasets spanning coding, science, engineering, and math across tens of thousands of NVIDIA GB300 GPUs.",
      "highlights": [
        "API model ID: grok-4.5, with a 500k-token context window",
        "Available at launch in Grok Build, Cursor on all plans, and the SpaceXSI console",
        "Rolled out to grok.com, X, iOS, and Android on July 22, 2026",
        "Configurable reasoning effort, default high"
      ],
      "source_title": "Introducing Grok 4.5 | SpaceXSI",
      "source_url": "https://x.ai/news/grok-4-5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#grok-4-5"
    },
    {
      "slug": "kimi-k3",
      "date": "2026-07-16",
      "model": "Kimi K3",
      "lab": "Moonshot AI",
      "lab_slug": "moonshot-ai",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1048576,
      "parameters": "MoE: 2.8T total, 104B active",
      "model_license": "Kimi K3 License",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "Kimi K3 is Moonshot AI's 2.8T-parameter model, which Moonshot describes as the first open 3T-class model. It is built on Kimi Delta Attention and Attention Residuals and targets long-horizon coding, knowledge work, and reasoning.",
      "highlights": [
        "Activates 16 of 896 experts per token",
        "Natively understands text, images, and video",
        "Launched on Kimi.com, Kimi Work, Kimi Code, and the Kimi API, with full weights promised by July 27, 2026",
        "Weights and code under the Kimi K3 License"
      ],
      "source_title": "Kimi K3: Open Frontier Intelligence",
      "source_url": "https://www.kimi.com/en/blog/kimi-k3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#kimi-k3"
    },
    {
      "slug": "minimax-music-3",
      "date": "2026-07-16",
      "model": "MiniMax Music 3.0",
      "lab": "MiniMax",
      "lab_slug": "minimax",
      "type": "music",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": "8B global LLM plus 0.6B local LLM",
      "model_license": "MiniMax-Music3 Community License",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "MiniMax Music 3.0 generates complete songs of up to five minutes from a creative description and optional lyrics, using a redesigned pipeline with an eight-layer RVQ tokenizer, a global-local Hybrid-LM, flow matching, and a Flow-VAE.",
      "highlights": [
        "Listed in the MiniMax API release notes on July 16, 2026",
        "Open weights published as MiniMax-Music3 with the August 13, 2026 announcement",
        "Produces 32 kHz, 16-bit stereo audio",
        "Music API closed to new users from August 20, 2026, with MiniMax pointing to MiniMax Audio and the open model"
      ],
      "source_title": "MiniMax Music 3.0: Next-Generation Open-Weights, Production-Ready & Versatile Music Model",
      "source_url": "https://www.minimax.io/blog/minimax-music-3-0-next-generation-open-weights-production-ready-versatile-music-model",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#minimax-music-3"
    },
    {
      "slug": "nemotron-3-embed",
      "date": "2026-07-16",
      "model": "Nemotron 3 Embed",
      "lab": "NVIDIA",
      "lab_slug": "nvidia",
      "type": "embedding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "embedding"
      ],
      "context_window": null,
      "parameters": "1B and 8B",
      "model_license": "OpenMDW-1.1",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Text embedding models for retrieval and semantic similarity with multilingual and cross-lingual retrieval, built with bidirectional attention and average pooling.",
      "highlights": [
        "Released in 1B and 8B sizes",
        "Intended as a foundation for RAG systems",
        "Released under the OpenMDW 1.1 license"
      ],
      "source_title": "nvidia/Nemotron-3-Embed-8B-BF16 · Hugging Face",
      "source_url": "https://huggingface.co/nvidia/Nemotron-3-Embed-8B-BF16",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#nemotron-3-embed"
    },
    {
      "slug": "hy-embodied-vlm-1-0",
      "date": "2026-07-15",
      "model": "Hy-Embodied-VLM-1.0",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "MoE: ~30B total, ~3B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Hy-Embodied-VLM-1.0 is an open mixture-of-experts vision-language foundation model for embodied agents, built on the Hy3-A3B language backbone and the Hy-ViT2 vision encoder for latency-sensitive physical-world perception and reasoning.",
      "highlights": [
        "About 30B total parameters with roughly 3B activated per token",
        "Activates about one-tenth the parameters of the previous A32B embodied system",
        "Weights on Hugging Face"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/HY-Embodied: HY-Embodied: Embodied Foundation Models for Real-World Agents",
      "source_url": "https://github.com/Tencent-Hunyuan/HY-Embodied",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#hy-embodied-vlm-1-0"
    },
    {
      "slug": "inkling",
      "date": "2026-07-15",
      "model": "Inkling",
      "lab": "Thinking Machines Lab",
      "lab_slug": "thinking-machines-lab",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": "MoE: 975B total, 41B active",
      "model_license": "Apache 2.0",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "The first model Thinking Machines Lab released: an open-weights mixture-of-experts transformer that reasons natively over text, images and audio, with controllable thinking effort, built to be customized through Tinker.",
      "highlights": [
        "975B total parameters, 41B active; pretrained on 45 trillion tokens of text, images, audio and video",
        "Context window up to 1M tokens; Tinker offered 64K and 256K options at launch",
        "Controllable thinking effort to trade performance against token use",
        "Weights on Hugging Face (original and NVFP4) under Apache 2.0; served at launch by Together AI, Fireworks, Modal, Databricks and Baseten"
      ],
      "source_title": "Inkling: Our Open-Weights Model",
      "source_url": "https://thinkingmachines.ai/news/introducing-inkling/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#inkling"
    },
    {
      "slug": "gpt-5-6",
      "date": "2026-07-09",
      "model": "GPT-5.6 (Sol, Terra, and Luna)",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1050000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API",
        "Enterprise"
      ],
      "summary": "GPT-5.6 is a model family with three durable tiers: Sol as the flagship, Terra balancing intelligence and cost, and Luna for efficient high-volume work, adding a max reasoning effort and an ultra multi-agent mode.",
      "highlights": [
        "API model IDs gpt-5.6-sol, gpt-5.6-terra, and gpt-5.6-luna; the gpt-5.6 alias routes to Sol",
        "General availability followed a limited June 26, 2026 preview for a small group of trusted partners",
        "In ChatGPT, Plus and higher plans use Sol, with GPT-5.6 Sol Pro for Pro and Enterprise users",
        "All three tiers list a 1,050,000-token context window on their API model pages"
      ],
      "source_title": "GPT-5.6: Frontier intelligence that scales with your ambition | OpenAI",
      "source_url": "https://openai.com/index/gpt-5-6/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#gpt-5-6"
    },
    {
      "slug": "muse-spark-1-1",
      "date": "2026-07-09",
      "model": "Muse Spark 1.1",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video",
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1048576,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Muse Spark 1.1 is a multimodal reasoning model built for agentic tasks, upgrading Muse Spark in tool and computer use, coding, multi-agent orchestration and context management.",
      "highlights": [
        "Available to developers through the Meta Model API public preview",
        "Manages a context window of 1 million tokens",
        "Available in Thinking mode in the Meta AI app and on meta.ai"
      ],
      "source_title": "Introducing Muse Spark 1.1",
      "source_url": "https://ai.meta.com/blog/introducing-muse-spark-meta-model-api/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#muse-spark-1-1"
    },
    {
      "slug": "gpt-live-1",
      "date": "2026-07-08",
      "model": "GPT-Live-1 and GPT-Live-1 mini",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio",
        "text"
      ],
      "output_modalities": [
        "audio",
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "GPT-Live is a new generation of full-duplex voice models that can listen and speak at the same time, delegating search and deeper reasoning to a frontier model in the background, and it now powers ChatGPT Voice.",
      "highlights": [
        "Rolled out to ChatGPT users globally: GPT-Live-1 for paid users and GPT-Live-1 mini for Free users",
        "At launch it delegated harder requests to GPT-5.5 in the background",
        "GPT-Live-1 became generally available in the API as gpt-live-1 on September 10, 2026",
        "Not available in ChatGPT Business, Enterprise, or Edu workspaces at launch"
      ],
      "source_title": "Introducing GPT-Live | OpenAI",
      "source_url": "https://openai.com/index/introducing-gpt-live/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#gpt-live-1"
    },
    {
      "slug": "seedream-5-0-pro",
      "date": "2026-07-08",
      "model": "Seedream 5.0 Pro",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Seedream 5.0 Pro is a multimodal image creation model aimed at professional design work, adding complex information visualization, interactive region-based editing, layer separation, and native multilingual text rendering to the Seedream line.",
      "highlights": [
        "Supports point, lasso, box, and sketch selections for precise local edits",
        "Can split a poster into more than 10 editable layers",
        "Generates text natively in over ten languages beyond Chinese and English",
        "Launched in Volcano Ark's experience center, rolling out to Doubao and Jimeng"
      ],
      "source_title": "Beyond Generation, It Understands Design | Introducing Seedream 5.0 Pro",
      "source_url": "https://seed.bytedance.com/en/blog/beyond-generation-it-understands-design-introducing-seedream-5-0-pro",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#seedream-5-0-pro"
    },
    {
      "slug": "cohere-transcribe-arabic",
      "date": "2026-07-07",
      "model": "Cohere Transcribe Arabic",
      "lab": "Cohere",
      "lab_slug": "cohere",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "2B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "Open-source speech-to-text model that fine-tunes Cohere Transcribe on Arabic speech data to transcribe all major Arabic dialects as well as English spoken with an Arabic accent.",
      "highlights": [
        "Conformer-based encoder-decoder architecture",
        "Model name cohere-transcribe-arabic-07-2026",
        "Open weights on Hugging Face"
      ],
      "source_title": "Meet Cohere Transcribe Arabic",
      "source_url": "https://docs.cohere.com/v2/changelog/transcribe-arabic",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#cohere-transcribe-arabic"
    },
    {
      "slug": "hunyuanocr-1-5",
      "date": "2026-07-07",
      "model": "HunyuanOCR-1.5",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Tencent Hunyuan Community License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "HunyuanOCR-1.5 upgrades Tencent's lightweight end-to-end OCR vision-language model without changing its backbone, adding DFlash speculative decoding, PC-side llama.cpp deployment, and an upgraded data and training recipe.",
      "highlights": [
        "Covers document parsing, text spotting, information extraction, and text-image translation",
        "Ships a DFlash block-diffusion draft model for faster decoding",
        "HunyuanOCR-1.0 weights remain archived in the same repository"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/HunyuanOCR: HunyuanOCR-1.5: Making Lightweight OCR VLMs Faster and Better",
      "source_url": "https://github.com/Tencent-Hunyuan/HunyuanOCR",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#hunyuanocr-1-5"
    },
    {
      "slug": "muse-image",
      "date": "2026-07-07",
      "model": "Muse Image",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Muse Image is Meta Superintelligence Labs' first image generation model, an agentic generator and editor that uses search and coding tools, refines its own outputs and composes from multiple references.",
      "highlights": [
        "Available in the Meta AI app, on meta.ai, and in Instagram Stories in the US at launch",
        "Supports multi-turn precise editing and multi-reference composition",
        "Outputs carry Content Seal, an invisible watermark",
        "Announced with Muse Video, which was only previewed"
      ],
      "source_title": "Introducing Muse Image and Muse Video",
      "source_url": "https://ai.meta.com/blog/introducing-muse-image-muse-video-msl/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#muse-image"
    },
    {
      "slug": "gpt-realtime-2-1",
      "date": "2026-07-06",
      "model": "gpt-realtime-2.1 and gpt-realtime-2.1-mini",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "audio",
        "image"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": 128000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "GPT-Realtime-2.1 updates GPT-Realtime-2 with improved alphanumeric recognition, silence and noise handling, and interruption behavior, and was released alongside GPT-Realtime-2.1 mini, a faster, lower-cost distilled reasoning model for realtime voice.",
      "highlights": [
        "API model IDs gpt-realtime-2.1 and gpt-realtime-2.1-mini in the Realtime API",
        "Both list a 128,000-token context window and 32,000 max output tokens",
        "The mini model supports audio and text input over WebRTC, WebSocket, or SIP connections"
      ],
      "source_title": "Changelog | OpenAI API",
      "source_url": "https://developers.openai.com/api/docs/changelog",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#gpt-realtime-2-1"
    },
    {
      "slug": "hy3",
      "date": "2026-07-06",
      "model": "Hy3",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "MoE: 295B total, 21B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Hy3 is the full release following Hy3 preview, with scaled-up post-training and reinforcement learning that improve reasoning, agentic, and long-context tasks, plus fixes for tool-call stability and hallucination based on feedback from over 50 products.",
      "highlights": [
        "295B total and 21B active parameters, plus 3.8B MTP layer parameters",
        "256K context length",
        "Weights on GitHub, Hugging Face, ModelScope, and AtomGit",
        "API on Tencent Cloud and OpenRouter"
      ],
      "source_title": "Introducing Hy3",
      "source_url": "https://hy.tencent.com/research/hy3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#hy3"
    },
    {
      "slug": "leanstral-1-5",
      "date": "2026-07-02",
      "model": "Leanstral 1.5",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "MoE: 119B total, 6B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Updated Lean 4 proof engineering model trained through mid-training, supervised fine-tuning and reinforcement learning with CISPO, aimed at automated theorem proving and code verification.",
      "highlights": [
        "API access began June 30, 2026 as labs-leanstral-1-5",
        "Trained in multi-turn proving and code agent environments",
        "Retired from the API September 30, 2026"
      ],
      "source_title": "Leanstral 1.5: Proof Abundance for All",
      "source_url": "https://mistral.ai/news/leanstral-1-5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/07#leanstral-1-5"
    },
    {
      "slug": "claude-sonnet-5",
      "date": "2026-06-30",
      "model": "Claude Sonnet 5",
      "lab": "Anthropic",
      "lab_slug": "anthropic",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Cloud platforms",
        "Enterprise"
      ],
      "summary": "Claude Sonnet 5 is the next generation of Anthropic's Sonnet line, built for agentic work such as planning, browser and terminal tool use, and autonomous runs, with performance Anthropic describes as close to Claude Opus 4.8.",
      "highlights": [
        "API model ID claude-sonnet-5; 1M-token context window and 128K max output tokens",
        "Default model for Claude Free and Pro plans",
        "Adaptive thinking on by default; manual extended thinking returns an error",
        "Updated tokenizer can map the same input to roughly 1.0 to 1.35 times as many tokens"
      ],
      "source_title": "Introducing Claude Sonnet 5",
      "source_url": "https://www.anthropic.com/news/claude-sonnet-5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#claude-sonnet-5"
    },
    {
      "slug": "longcat-2-0",
      "date": "2026-06-30",
      "model": "LongCat-2.0",
      "lab": "Meituan",
      "lab_slug": "meituan",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": "MoE: 1.6T total, about 48B active",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "LongCat-2.0 is Meituan's large mixture-of-experts language model for coding and agentic work, trained entirely on AI ASIC superpods and introducing LongCat Sparse Attention for efficient long-context processing.",
      "highlights": [
        "1.6 trillion total parameters with about 48B activated per token",
        "Trained on 1M-context data using LongCat Sparse Attention",
        "Pretrained on more than 35 trillion tokens",
        "Weights released under MIT; API billing launched the same day"
      ],
      "source_title": "Introducing LongCat-2.0",
      "source_url": "https://longcat.chat/blog/longcat-2.0/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#longcat-2-0"
    },
    {
      "slug": "nano-banana-2-lite",
      "date": "2026-06-30",
      "model": "Nano Banana 2 Lite (Gemini 3.1 Flash-Lite Image)",
      "lab": "Google",
      "lab_slug": "google",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text",
        "image"
      ],
      "context_window": 65536,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms",
        "Consumer app"
      ],
      "summary": "Nano Banana 2 Lite (Gemini 3.1 Flash-Lite Image) is the fastest and most cost-efficient model in the Nano Banana family, built for high-throughput image generation and editing.",
      "highlights": [
        "Generally available at launch in Google AI Studio, the Gemini API and Gemini Enterprise Agent Platform",
        "Recommended replacement for the original Nano Banana (gemini-2.5-flash-image)",
        "Also rolled out in AI Mode in Search and the Gemini app"
      ],
      "source_title": "Start building with Nano Banana 2 Lite and Gemini Omni Flash",
      "source_url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-omni-flash-nano-banana-2-lite/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#nano-banana-2-lite"
    },
    {
      "slug": "lfm2-5-230m",
      "date": "2026-06-25",
      "model": "LFM2.5-230M",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32768,
      "parameters": "230M",
      "model_license": "LFM Open License v1.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Liquid AI's smallest model at release, a 230M-parameter LFM2-architecture model pre-trained on 19T tokens with a 32K context extension and a light post-training recipe meant as a base for fine-tuning in agentic workflows.",
      "highlights": [
        "Released as LFM2.5-230M-Base and post-trained LFM2.5-230M",
        "Post-training: distillation from LFM2.5-350M, DPO, and multi-domain RL",
        "Demonstrated as a skill-selection layer on a Unitree G1 humanoid robot",
        "Available on Hugging Face"
      ],
      "source_title": "LFM2.5-230M: Built to Run Anywhere",
      "source_url": "https://www.liquid.ai/blog/lfm2-5-230m",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#lfm2-5-230m"
    },
    {
      "slug": "mistral-ocr-4",
      "date": "2026-06-23",
      "model": "Mistral OCR 4",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Enterprise"
      ],
      "summary": "OCR model that returns bounding boxes, typed block classification and inline confidence scores with extracted text, supporting 170 languages and single-container self-hosting.",
      "highlights": [
        "Ingestion component of Mistral's Search Toolkit",
        "OCR 4.1 followed on July 16, 2026 with block-level confidence scores",
        "API model ID mistral-ocr-4-0"
      ],
      "source_title": "Mistral OCR 4 : SOTA OCR for Document Intelligence",
      "source_url": "https://mistral.ai/news/ocr-4",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#mistral-ocr-4"
    },
    {
      "slug": "seed-2-1",
      "date": "2026-06-23",
      "model": "Seed2.1",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Seed2.1 is ByteDance Seed's model family for agentic productivity, offered in Pro and Turbo sizes, with more reliable multi-step agent execution, end-to-end coding delivery, and stronger understanding of complex documents and video than Seed2.0.",
      "highlights": [
        "Released in Pro and Turbo sizes",
        "Available to Doubao app and Volcano Engine users at launch",
        "Volcano Engine's June 2026 release notes also add doubao-seed-evolving, a Seed model updated at least weekly"
      ],
      "source_title": "Seed2.1 Officially Released: Advancing AI Productivity",
      "source_url": "https://seed.bytedance.com/en/blog/seed2-1-officially-released-advancing-ai-productivity",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#seed-2-1"
    },
    {
      "slug": "baidu-unlimited-ocr",
      "date": "2026-06-22",
      "model": "Unlimited-OCR",
      "lab": "Baidu",
      "lab_slug": "baidu",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "image",
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Unlimited-OCR is an open vision-language OCR model from Baidu aimed at one-shot, long-horizon parsing of documents, which Baidu describes as pushing the DeepSeek-OCR approach one step further.",
      "highlights": [
        "Weights on Hugging Face and ModelScope under the MIT license",
        "vLLM inference support added June 28, 2026",
        "Made available on Baidu Cloud on July 3, 2026"
      ],
      "source_title": "baidu/Unlimited-OCR · Hugging Face",
      "source_url": "https://huggingface.co/baidu/Unlimited-OCR",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#baidu-unlimited-ocr"
    },
    {
      "slug": "lfm2-5-retrievers",
      "date": "2026-06-18",
      "model": "LFM2.5-Embedding-350M and LFM2.5-ColBERT-350M",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "embedding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "embedding"
      ],
      "context_window": null,
      "parameters": "350M each",
      "model_license": "LFM Open License v1.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Two multilingual retrieval models, the first bidirectional members of the LFM family: a single-vector embedding model and a late-interaction ColBERT model, both 350M parameters and built from LFM2.5-350M-Base.",
      "highlights": [
        "Supports 11 languages for multilingual and cross-lingual search",
        "Causal attention and short convolutions converted to bidirectional versions",
        "Positioned for short-context search; 512-token document length per model card",
        "Available on Hugging Face"
      ],
      "source_title": "LFM2.5 Retrievers: Bi-directional LFMs for Fast Multilingual Search",
      "source_url": "https://www.liquid.ai/blog/lfm2-5-retrievers",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#lfm2-5-retrievers"
    },
    {
      "slug": "kling-3-0-turbo",
      "date": "2026-06-17",
      "model": "Kling 3.0 Turbo",
      "lab": "Kuaishou",
      "lab_slug": "kuaishou",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Kling 3.0 Turbo is a new model in Kuaishou's Kling 3.0 video series, launched as part of a series upgrade that also improved video editing in Kling 3.0 Omni.",
      "highlights": [
        "Listed as a Kling 3.0 model series upgrade in Kling's release notes",
        "Launched alongside upgraded video editing in Kling 3.0 Omni"
      ],
      "source_title": "Kling AI Release Notes & Version History | Kling AI",
      "source_url": "https://kling.ai/release-note/release-history",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#kling-3-0-turbo"
    },
    {
      "slug": "glm-5-2",
      "date": "2026-06-16",
      "model": "GLM-5.2",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": "MoE: 744B total, 40B active",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "GLM-5.2 is Z.ai's flagship model for long-horizon tasks and the first GLM model to deliver its capabilities on a 1M-token context. It adds multiple thinking effort levels for balancing coding performance and latency.",
      "highlights": [
        "IndexShare reuses one indexer across every four sparse attention layers, cutting per-token FLOPs at 1M context",
        "Improved multi-token prediction layer for speculative decoding",
        "Up to 128K output tokens",
        "Weights under the MIT license"
      ],
      "source_title": "GLM-5.2: Built for Long-Horizon Tasks",
      "source_url": "https://z.ai/blog/glm-5.2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#glm-5-2"
    },
    {
      "slug": "kimi-k2-7-code",
      "date": "2026-06-12",
      "model": "Kimi K2.7 Code",
      "lab": "Moonshot AI",
      "lab_slug": "moonshot-ai",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "MoE: 1T total, 32B active",
      "model_license": "Modified MIT License",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "Kimi K2.7 Code is a coding-focused agentic model built on Kimi K2.6, with improvements on real-world long-horizon coding tasks. Moonshot says it uses about 30% fewer thinking tokens than K2.6.",
      "highlights": [
        "Released through the Kimi Open Platform API, followed by a HighSpeed variant",
        "MoonViT vision encoder with 400M parameters",
        "Weights on Hugging Face under the Modified MIT License"
      ],
      "source_title": "moonshotai/Kimi-K2.7-Code (Hugging Face)",
      "source_url": "https://huggingface.co/moonshotai/Kimi-K2.7-Code",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#kimi-k2-7-code"
    },
    {
      "slug": "claude-fable-5",
      "date": "2026-06-09",
      "model": "Claude Fable 5",
      "lab": "Anthropic",
      "lab_slug": "anthropic",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Cloud platforms",
        "Enterprise"
      ],
      "summary": "Claude Fable 5 is a Mythos-class Anthropic model released with safeguards for general use, aimed at long, complex software engineering, knowledge work, and research tasks. Requests on some sensitive topics are answered by Claude Opus 4.8 instead.",
      "highlights": [
        "API model ID claude-fable-5; 1M-token context window and 128K max output tokens",
        "Claude Mythos 5, the same model with some safeguards lifted, launched only for Project Glasswing partners",
        "Access was suspended June 12, 2026 under a US export control directive and restored July 1, 2026",
        "Adaptive thinking is the only thinking mode"
      ],
      "source_title": "Claude Fable 5 and Claude Mythos 5",
      "source_url": "https://www.anthropic.com/news/claude-fable-5-mythos-5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#claude-fable-5"
    },
    {
      "slug": "gemini-3-5-live-translate",
      "date": "2026-06-09",
      "model": "Gemini 3.5 Live Translate",
      "lab": "Google",
      "lab_slug": "google",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "audio",
        "text"
      ],
      "context_window": 131072,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Enterprise"
      ],
      "summary": "Gemini 3.5 Live Translate is an audio model for near real-time speech-to-speech translation that detects more than 70 languages and preserves the speaker's intonation, pacing and pitch.",
      "highlights": [
        "Public preview for developers in the Gemini Live API and Google AI Studio",
        "Private preview for select Google Workspace business customers in Google Meet",
        "Generates translated speech continuously instead of waiting for the speaker to finish"
      ],
      "source_title": "Gemini 3.5 Live Translate is here",
      "source_url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-live-3-5-translate/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#gemini-3-5-live-translate"
    },
    {
      "slug": "north-mini-code",
      "date": "2026-06-09",
      "model": "North Mini Code",
      "lab": "Cohere",
      "lab_slug": "cohere",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "MoE: 30B total, 3B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "Cohere's first agentic coding model, a mixture-of-experts model trained specifically for agentic coding whose small active parameter footprint lets it run on local hardware.",
      "highlights": [
        "256K input and 64K output tokens",
        "Available through the Chat V2 API and as open weights",
        "Supported in Cohere's Model Vault"
      ],
      "source_title": "Announcing Cohere's North Mini Code",
      "source_url": "https://docs.cohere.com/v2/changelog/north-mini-code-1-0",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#north-mini-code"
    },
    {
      "slug": "luma-ray3-2",
      "date": "2026-06-09",
      "model": "Ray3.2",
      "lab": "Luma AI",
      "lab_slug": "luma-ai",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Ray3 update designed with entertainment, advertising and gaming creatives that adds frame-level direction and, for the first time, exposes the full Ray control surface through an API.",
      "highlights": [
        "Multi-keyframe direction with up to 16 keyframes",
        "Native HDR generation with EXR output",
        "Modify Video V2 up to 20 seconds at 1080p"
      ],
      "source_title": "Luma Introduces Ray3.2 Model & API: Complete Creative Control for Video Generation",
      "source_url": "https://lumalabs.ai/news/introducing-ray-3-2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#luma-ray3-2"
    },
    {
      "slug": "scail-2",
      "date": "2026-06-09",
      "model": "SCAIL-2",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "SCAIL-2 is Z.ai's open-weights character animation model, which unifies controlled character animation with replacement in one video generation framework. It succeeds the SCAIL-Preview research release.",
      "highlights": [
        "Model and inference code released on June 9, 2026",
        "Multi-reference inference added on June 13, 2026",
        "Weights on Hugging Face under the MIT license"
      ],
      "source_title": "SCAIL-2 (GitHub)",
      "source_url": "https://github.com/zai-org/SCAIL-2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#scail-2"
    },
    {
      "slug": "apple-adm-3-cloud",
      "date": "2026-06-08",
      "model": "ADM 3 Cloud",
      "lab": "Apple",
      "lab_slug": "apple",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "ADM 3 Cloud is Apple's third-generation server image model for generating and editing images, powering photo editing tools, the new Image Playground and Genmoji.",
      "highlights": [
        "Runs on Private Cloud Compute",
        "Specialized adapters support features such as Spatial Reframing in Photos"
      ],
      "source_title": "Introducing the Third Generation of Apple's Foundation Models",
      "source_url": "https://machinelearning.apple.com/research/introducing-third-generation-of-apple-foundation-models",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#apple-adm-3-cloud"
    },
    {
      "slug": "apple-afm-3-cloud",
      "date": "2026-06-08",
      "model": "AFM 3 Cloud",
      "lab": "Apple",
      "lab_slug": "apple",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "AFM 3 Cloud is the server-side workhorse of Apple's third-generation foundation models, optimized for speed and efficiency with multimodal reasoning on Private Cloud Compute.",
      "highlights": [
        "Runs on Private Cloud Compute on Apple silicon servers",
        "Apple says user data is never stored or shared, including with Apple"
      ],
      "source_title": "Introducing the Third Generation of Apple's Foundation Models",
      "source_url": "https://machinelearning.apple.com/research/introducing-third-generation-of-apple-foundation-models",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#apple-afm-3-cloud"
    },
    {
      "slug": "apple-afm-3-cloud-pro",
      "date": "2026-06-08",
      "model": "AFM 3 Cloud Pro",
      "lab": "Apple",
      "lab_slug": "apple",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "AFM 3 Cloud Pro is Apple's most capable server-based foundation model in the third generation, powering demanding Apple Intelligence uses such as agentic tool use and complex reasoning.",
      "highlights": [
        "Runs on Private Cloud Compute extended to NVIDIA GPUs in Google Cloud",
        "Part of a model family custom-built in collaboration with Google"
      ],
      "source_title": "Introducing the Third Generation of Apple's Foundation Models",
      "source_url": "https://machinelearning.apple.com/research/introducing-third-generation-of-apple-foundation-models",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#apple-afm-3-cloud-pro"
    },
    {
      "slug": "apple-afm-3-core",
      "date": "2026-06-08",
      "model": "AFM 3 Core",
      "lab": "Apple",
      "lab_slug": "apple",
      "type": "small",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "3B (dense)",
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "AFM 3 Core is the third-generation version of Apple's dense 3B on-device foundation model, a step up in quality optimized for Apple silicon and part of a family built with Google.",
      "highlights": [
        "Successor to Apple's 3B dense on-device model",
        "Image understanding evaluated alongside text",
        "Third-generation family custom-built in collaboration with Google"
      ],
      "source_title": "Introducing the Third Generation of Apple's Foundation Models",
      "source_url": "https://machinelearning.apple.com/research/introducing-third-generation-of-apple-foundation-models",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#apple-afm-3-core"
    },
    {
      "slug": "apple-afm-3-core-advanced",
      "date": "2026-06-08",
      "model": "AFM 3 Core Advanced",
      "lab": "Apple",
      "lab_slug": "apple",
      "type": "small",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": null,
      "parameters": "Sparse: 20B total, 1B to 4B active",
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "AFM 3 Core Advanced is Apple's most powerful on-device model, a natively multimodal sparse model that supports expressive voices and higher-accuracy dictation on Apple's most capable devices.",
      "highlights": [
        "Activates 1 to 4 billion of its 20 billion parameters depending on the request",
        "Keeps full weights in flash memory and loads active experts into DRAM as needed",
        "Limited to Apple's most capable Apple silicon systems"
      ],
      "source_title": "Introducing the Third Generation of Apple's Foundation Models",
      "source_url": "https://machinelearning.apple.com/research/introducing-third-generation-of-apple-foundation-models",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#apple-afm-3-core-advanced"
    },
    {
      "slug": "magenta-realtime-2",
      "date": "2026-06-04",
      "model": "Magenta RealTime 2",
      "lab": "Google",
      "lab_slug": "google",
      "type": "music",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": "2.4B (base), 230M (small)",
      "model_license": "CC-BY-4.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Magenta RealTime 2 is an open-weights live music model from Google DeepMind's Magenta team for real-time, interactive generation on a laptop. It follows Magenta RealTime with frame-wise generation, MIDI control and much lower control latency.",
      "highlights": [
        "Base model has 2.4B parameters and small model 230M",
        "Control latency of about 200 ms versus about 3 seconds for the original",
        "MLX-based C++ engine runs natively on Apple Silicon",
        "Weights under CC BY 4.0, code under Apache 2.0"
      ],
      "source_title": "Magenta RealTime 2: Open & Local Live Music Models",
      "source_url": "https://magenta.withgoogle.com/magenta-realtime-2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#magenta-realtime-2"
    },
    {
      "slug": "nemotron-3-ultra",
      "date": "2026-06-04",
      "model": "Nemotron 3 Ultra",
      "lab": "NVIDIA",
      "lab_slug": "nvidia",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": "MoE: 550B total, 55B active",
      "model_license": "OpenMDW-1.1",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Frontier-scale Nemotron 3 model for complex agentic workflows, long-context analysis and multilingual reasoning, built on a hybrid latent mixture-of-experts design.",
      "highlights": [
        "Supports up to 1M tokens of context",
        "Includes multi-token prediction layers",
        "Released under the OpenMDW 1.1 license"
      ],
      "source_title": "nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 · Hugging Face",
      "source_url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#nemotron-3-ultra"
    },
    {
      "slug": "gemma-4-12b",
      "date": "2026-06-03",
      "model": "Gemma 4 12B",
      "lab": "Google",
      "lab_slug": "google",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 262144,
      "parameters": "12B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "Gemma 4 12B is a mid-sized, encoder-free multimodal Gemma 4 model in which vision and audio inputs flow directly into the language model backbone. It sits between the E4B and 26B MoE models and targets laptops.",
      "highlights": [
        "No separate vision or audio encoders; inputs are projected into the LLM embedding space",
        "First mid-sized Gemma model with native audio input",
        "Runs locally with 16GB of VRAM or unified memory",
        "256K-token context per the Hugging Face model card; ships with Multi-Token Prediction drafters"
      ],
      "source_title": "Introducing Gemma 4 12B: a unified, encoder-free multimodal model",
      "source_url": "https://blog.google/innovation-and-ai/technology/developers-tools/introducing-gemma-4-12b/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#gemma-4-12b"
    },
    {
      "slug": "gpt-rosalind-5-5",
      "date": "2026-06-03",
      "model": "GPT-Rosalind-5.5",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [],
      "output_modalities": [],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Enterprise"
      ],
      "summary": "GPT-Rosalind-5.5 updates OpenAI's GPT-Rosalind life sciences series, combining GPT-5.5's agentic coding and tool-use capabilities with stronger model intelligence in drug-discovery domains such as medicinal chemistry and genomics.",
      "highlights": [
        "With this release, OpenAI expanded access to the GPT-Rosalind series to eligible organizations globally through trusted access",
        "Added Codex plugins for evidence retrieval, biological interpretation, and bioinformatics, open to all users and powered by GPT-Rosalind for qualified enterprise users",
        "Added interactive Codex viewers for sequence, alignment, and structure files"
      ],
      "source_title": "Introducing GPT-Rosalind-5.5 | OpenAI",
      "source_url": "https://openai.com/index/introducing-new-capabilities-to-gpt-rosalind/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#gpt-rosalind-5-5"
    },
    {
      "slug": "grok-imagine-video-1-5",
      "date": "2026-06-03",
      "model": "Grok Imagine Video 1.5",
      "lab": "xAI",
      "lab_slug": "xai",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video",
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Grok Imagine Video 1.5 is SpaceXSI's image-to-video model with generated audio. It launched in API preview, then became generally available with improved motion, physics, and synced sound and speech.",
      "highlights": [
        "Preview model ID grok-imagine-video-1.5-preview, with clips up to 720p",
        "GA as grok-imagine-video-1.5 on June 16, 2026; Video 1.5 Fast added to grok.com and apps",
        "July 31, 2026 update added image and voice references, text-to-video, and native 1080p"
      ],
      "source_title": "Grok Imagine 1.5 Preview | SpaceXSI",
      "source_url": "https://x.ai/news/grok-imagine-1-5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#grok-imagine-video-1-5"
    },
    {
      "slug": "ideogram-4-0",
      "date": "2026-06-03",
      "model": "Ideogram 4.0",
      "lab": "Ideogram",
      "lab_slug": "ideogram",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "9.3B",
      "model_license": "Ideogram 4 Non-Commercial License",
      "available_at_launch": [
        "Open weights",
        "Consumer app",
        "API"
      ],
      "summary": "Ideogram's first open-weight foundation model, a 9.3B-parameter single-stream diffusion transformer trained from scratch with a vision-language text encoder and structured JSON prompts.",
      "highlights": [
        "9.3B-parameter DiT trained from scratch",
        "Weights on Hugging Face, code on GitHub",
        "Production use requires a commercial license"
      ],
      "source_title": "Ideogram 4.0 Technical Details: Open model at the forefront of design",
      "source_url": "https://ideogram.ai/blog/ideogram-4.0/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#ideogram-4-0"
    },
    {
      "slug": "mai-code-1-flash",
      "date": "2026-06-02",
      "model": "MAI-Code-1-Flash",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "coding",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "5B active",
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "MAI-Code-1-Flash is Microsoft AI's inference-efficient agentic coding model, trained inside the agent tooling GitHub Copilot uses and rolled out to GitHub Copilot users in Visual Studio Code. It uses adaptive thinking to scale reasoning to task difficulty.",
      "highlights": [
        "Rolled out to GitHub Copilot individual users in VS Code",
        "Has 5 billion active parameters",
        "Trained without distillation from third-party models"
      ],
      "source_title": "Introducing MAI-Code-1-Flash",
      "source_url": "https://microsoft.ai/news/introducingmai-code-1-flash/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#mai-code-1-flash"
    },
    {
      "slug": "mai-image-2-5",
      "date": "2026-06-02",
      "model": "MAI-Image-2.5",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Enterprise"
      ],
      "summary": "MAI-Image-2.5 is Microsoft AI's image model for high-quality text-to-image generation and precise, localized image editing, with face and identity consistency across edits. It launched in Foundry and in PowerPoint.",
      "highlights": [
        "Available to developers in Microsoft Foundry",
        "Live in PowerPoint and rolling out to OneDrive for photo editing",
        "Launched with a faster MAI-Image-2.5-Flash variant"
      ],
      "source_title": "MAI-Image-2.5 launches at No. 2 for image editing on Arena",
      "source_url": "https://microsoft.ai/news/introducing-mai-image-2-5/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#mai-image-2-5"
    },
    {
      "slug": "mai-image-2-5-flash",
      "date": "2026-06-02",
      "model": "MAI-Image-2.5-Flash",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "MAI-Image-2.5-Flash is a faster, lower-cost variant of Microsoft AI's MAI-Image-2.5 for scalable production image generation and editing workloads, launched in Microsoft Foundry on the same day as MAI-Image-2.5.",
      "highlights": [
        "Supports both text-to-image generation and image editing",
        "Available in Microsoft Foundry",
        "Part of the seven-model MAI launch at Microsoft Build 2026"
      ],
      "source_title": "MAI-Image-2.5 launches at No. 2 for image editing on Arena",
      "source_url": "https://microsoft.ai/news/introducing-mai-image-2-5/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#mai-image-2-5-flash"
    },
    {
      "slug": "mai-thinking-1",
      "date": "2026-06-02",
      "model": "MAI-Thinking-1",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "MoE: about 1T total, 35B active",
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "MAI-Thinking-1 is Microsoft AI's flagship reasoning model, a medium-sized sparse mixture-of-experts model trained from scratch without distillation from third-party models, aimed at software engineering, math and enterprise tasks.",
      "highlights": [
        "One of seven in-house models Microsoft AI announced on June 2, 2026",
        "Public preview in Microsoft Foundry from August 12, 2026",
        "Sparse mixture-of-experts with about 1T total and 35B active parameters"
      ],
      "source_title": "Building a hill-climbing machine: Launching seven new MAI models",
      "source_url": "https://microsoft.ai/news/building-a-hillclimbing-machine-launching-seven-new-mai-models/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#mai-thinking-1"
    },
    {
      "slug": "mai-transcribe-1-5",
      "date": "2026-06-02",
      "model": "MAI-Transcribe-1.5",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Enterprise"
      ],
      "summary": "MAI-Transcribe-1.5 is Microsoft AI's multilingual speech-to-text model, expanding language coverage from 25 to 43 languages and adding keyword biasing for domain-specific terminology, and is being integrated into Copilot, Teams, GitHub and Dynamics 365.",
      "highlights": [
        "Covers 43 languages, up from 25 in MAI-Transcribe-1",
        "Adds keyword biasing for user-supplied domain terms",
        "Available in Microsoft Foundry"
      ],
      "source_title": "Introducing MAI-Transcribe-1.5",
      "source_url": "https://microsoft.ai/news/mai-transcribe-1-5more-accurate-context-aware-and-built-for-production/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#mai-transcribe-1-5"
    },
    {
      "slug": "mai-voice-2",
      "date": "2026-06-02",
      "model": "MAI-Voice-2",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "MAI-Voice-2 is Microsoft AI's text-to-speech model, expanding from English-only to 15 languages with emotion tags and zero-shot voice prompting from short reference audio, protected by consent guardrails.",
      "highlights": [
        "Supports 15 languages",
        "Zero-shot voice prompting from 5 to 60 seconds of reference audio",
        "Available in Microsoft Foundry and being integrated into VS Code and Dynamics 365 Contact Center"
      ],
      "source_title": "Introducing MAI-Voice-2",
      "source_url": "https://microsoft.ai/news/mai-voice-2/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#mai-voice-2"
    },
    {
      "slug": "minimax-m3",
      "date": "2026-06-01",
      "model": "MiniMax M3",
      "lab": "MiniMax",
      "lab_slug": "minimax",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": "MoE: about 428B total, 23B active",
      "model_license": "MiniMax Community License",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "MiniMax M3 is a natively multimodal open-weight model aimed at coding and agentic work. It introduces MiniMax Sparse Attention (MSA) to reach a context window of up to 1M tokens and accepts image and video input.",
      "highlights": [
        "About 428B total parameters with about 23B activated, per the model card",
        "Supports three reasoning modes through a thinking parameter",
        "Can operate a desktop computer, enabling computer use in MiniMax Code",
        "MSA is open-sourced as a separate GitHub repository"
      ],
      "source_title": "MiniMax M3: Frontier Coding, 1M Context, Native Multimodality, All in One Model",
      "source_url": "https://www.minimax.io/blog/minimax-m3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#minimax-m3"
    },
    {
      "slug": "qwen3-7-plus",
      "date": "2026-06-01",
      "model": "Qwen3.7-Plus",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Qwen3.7-Plus is a multimodal agent model that unifies vision and language, building on the Qwen3.7 text backbone. It can read screens, operate GUIs, write code from visual references, and navigate mobile apps.",
      "highlights": [
        "Accepts text, image, and video input",
        "Supports preserving thinking content across turns for agentic tasks",
        "Available through Alibaba Cloud Model Studio"
      ],
      "source_title": "Qwen3.7-Plus: Multimodal Agent Intelligence",
      "source_url": "https://qwen.ai/blog?id=qwen3.7-plus",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/06#qwen3-7-plus"
    },
    {
      "slug": "cosmos-3",
      "date": "2026-05-31",
      "model": "Cosmos 3",
      "lab": "NVIDIA",
      "lab_slug": "nvidia",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "video",
        "image",
        "audio"
      ],
      "context_window": null,
      "parameters": "Nano 16B; Super 64B",
      "model_license": "OpenMDW-1.1",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Omnimodal world models that generate video, images, audio and robot action commands from text, image, video and action trajectory inputs for physical AI.",
      "highlights": [
        "Released on Hugging Face and GitHub",
        "Cosmos3-Edge, a 4B model, followed on July 20, 2026",
        "Released under the OpenMDW 1.1 license"
      ],
      "source_title": "nvidia/Cosmos3-Super · Hugging Face",
      "source_url": "https://huggingface.co/nvidia/Cosmos3-Super",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/05#cosmos-3"
    },
    {
      "slug": "grok-build-0-1",
      "date": "2026-05-29",
      "model": "Grok Build 0.1",
      "lab": "xAI",
      "lab_slug": "xai",
      "type": "coding",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Grok Build 0.1 is SpaceXSI's coding model trained for agentic coding tasks such as web development, debugging, and MCP support. It is the same model that powers the Grok Build coding agent, released on the API in public beta.",
      "highlights": [
        "API model ID: grok-build-0.1",
        "256k-token context window per the docs model page",
        "Served at 100+ tokens per second",
        "Docs list grok-code-fast-1 as an alias of grok-build-0.1"
      ],
      "source_title": "Grok Build 0.1 on API | SpaceXSI",
      "source_url": "https://x.ai/news/grok-build-0-1",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/05#grok-build-0-1"
    },
    {
      "slug": "step-3-7-flash",
      "date": "2026-05-29",
      "model": "Step 3.7 Flash",
      "lab": "StepFun",
      "lab_slug": "stepfun",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "MoE: 198B total, about 11B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "Step 3.7 Flash is a sparse mixture-of-experts vision-language model that builds on Step 3.5 Flash, adding native visual input for agent workflows that combine perception, search, coding, and reasoning at high throughput.",
      "highlights": [
        "196B-parameter language backbone plus a 1.8B-parameter vision encoder",
        "Activates about 11B parameters per token with a 256K context window",
        "Three selectable reasoning levels: low, medium, and high",
        "Weights released under Apache-2.0"
      ],
      "source_title": "Step 3.7 Flash: A high-efficiency Flash model for Real-World Agents",
      "source_url": "https://static.stepfun.com/blog/step-3.7-flash/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/05#step-3-7-flash"
    },
    {
      "slug": "claude-opus-4-8",
      "date": "2026-05-28",
      "model": "Claude Opus 4.8",
      "lab": "Anthropic",
      "lab_slug": "anthropic",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Cloud platforms"
      ],
      "summary": "Claude Opus 4.8 is an upgrade to Claude Opus 4.7 with improvements in coding, agentic tasks, reasoning, and practical knowledge work. It launched alongside effort controls in claude.ai and a dynamic workflows research preview in Claude Code.",
      "highlights": [
        "API model ID claude-opus-4-8",
        "1M-token context window by default on the Claude API, Amazon Bedrock, Google Cloud, and Microsoft Foundry",
        "128K max output tokens",
        "Supports mid-conversation system messages and high-resolution image input"
      ],
      "source_title": "Introducing Claude Opus 4.8",
      "source_url": "https://www.anthropic.com/news/claude-opus-4-8",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/05#claude-opus-4-8"
    },
    {
      "slug": "lfm2-5-8b-a1b",
      "date": "2026-05-28",
      "model": "LFM2.5-8B-A1B",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "MoE: 8.3B total, 1.5B active",
      "model_license": "LFM Open License v1.0",
      "available_at_launch": [
        "Consumer app",
        "Open weights"
      ],
      "summary": "An updated on-device mixture-of-experts model built for tool calling, expanding LFM2-8B-A1B with a 128K context window, pretraining scaled from 12T to 38T tokens, a doubled vocabulary, and reasoning-only output.",
      "highlights": [
        "Context window expanded from 32,768 to 128,000 tokens",
        "Vocabulary grows from 65,536 to 128,000 tokens for non-Latin scripts",
        "Produces an explicit chain of thought before every final answer",
        "Base and post-trained models on Hugging Face; testable on Liquid Playground"
      ],
      "source_title": "LFM2.5-8B-A1B: An Even Better On-Device Mixture of Experts",
      "source_url": "https://www.liquid.ai/blog/lfm2-5-8b-a1b",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/05#lfm2-5-8b-a1b"
    },
    {
      "slug": "eleven-music-v2",
      "date": "2026-05-26",
      "model": "Music v2",
      "lab": "ElevenLabs",
      "lab_slug": "elevenlabs",
      "type": "music",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Second-generation ElevenLabs music model with better vocals, instrumentation and arrangement across genres and improved multilingual support, powering ElevenMusic, ElevenAPI and ElevenCreative.",
      "highlights": [
        "Model ID music_v2",
        "Generates from prompts, composition plans and earlier songs",
        "Reference-based sound control added later"
      ],
      "source_title": "Introducing Music v2",
      "source_url": "https://elevenlabs.io/blog/introducing-music-v2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/05#eleven-music-v2"
    },
    {
      "slug": "keye-vl-2-0",
      "date": "2026-05-25",
      "model": "Keye-VL-2.0-30B-A3B",
      "lab": "Kuaishou",
      "lab_slug": "kuaishou",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "MoE: 30B total, 3B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Keye-VL-2.0-30B-A3B is the 30B-class flagship of Kuaishou's Keye multimodal series, using a DSA sparse-attention architecture to support long-video understanding over a 256K context.",
      "highlights": [
        "Sparse attention architecture for hour-long video understanding",
        "Nearly lossless reasoning over 256K context, per the model card",
        "Weights released under Apache-2.0"
      ],
      "source_title": "GitHub - Kwai-Keye/Keye",
      "source_url": "https://github.com/Kwai-Keye/Keye",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/05#keye-vl-2-0"
    },
    {
      "slug": "runway-aleph-2",
      "date": "2026-05-21",
      "model": "Aleph 2.0",
      "lab": "Runway",
      "lab_slug": "runway",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Upgrade to Runway's flagship video editing model, launched with Edit Studio, that edits clips up to 30 seconds at 1080p with localized changes and image-level control from an edited frame.",
      "highlights": [
        "Edits clips up to 30 seconds at 1080p",
        "Applies a user-edited frame across the whole video",
        "Added to Runway MCP in July 2026"
      ],
      "source_title": "Introducing Aleph 2.0 and Edit Studio",
      "source_url": "https://runway.com/news/introducing-aleph-2-and-edit-studio",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/05#runway-aleph-2"
    },
    {
      "slug": "fara-1-5",
      "date": "2026-05-21",
      "model": "Fara1.5",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 262144,
      "parameters": "4B, 9B, 27B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Fara1.5 is Microsoft's second generation of open computer use agent models, released in 4B, 9B and 27B sizes and fine-tuned from Qwen3.5 on synthetic web-task trajectories generated by the FaraGen1.5 pipeline.",
      "highlights": [
        "Released in 4B, 9B and 27B sizes with a 262,144-token context length",
        "Supervised fine-tuned from Qwen3.5 base models",
        "27B version also offered on Microsoft Foundry"
      ],
      "source_title": "microsoft/Fara1.5-27B on Hugging Face",
      "source_url": "https://huggingface.co/microsoft/Fara1.5-27B",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/05#fara-1-5"
    },
    {
      "slug": "hy-mt2",
      "date": "2026-05-21",
      "model": "Hy-MT2",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "1.8B, 7B, and MoE: 30B total, 3B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Hy-MT2 is a family of open fast-thinking multilingual translation models in 1.8B, 7B, and 30B-A3B mixture-of-experts sizes, supporting translation among 33 languages and following translation instructions given in multiple languages.",
      "highlights": [
        "Three sizes: 1.8B, 7B, and 30B-A3B (MoE)",
        "Released with IFMTBench, a benchmark for translation instruction following",
        "FP8 and GGUF versions published on Hugging Face"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/Hy-MT2",
      "source_url": "https://github.com/Tencent-Hunyuan/Hy-MT2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/05#hy-mt2"
    },
    {
      "slug": "longcat-video-avatar-1-5",
      "date": "2026-05-21",
      "model": "LongCat-Video-Avatar 1.5",
      "lab": "Meituan",
      "lab_slug": "meituan",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "audio"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "LongCat-Video-Avatar 1.5 is an upgraded open-source framework for audio-driven human video generation from Meituan's LongCat team, building on the LongCat-Video foundation model.",
      "highlights": [
        "Generates talking and expressive character video from audio",
        "Supports single- and multi-speaker audio-to-video generation",
        "Weights released under MIT"
      ],
      "source_title": "GitHub - meituan-longcat/LongCat-Video",
      "source_url": "https://github.com/meituan-longcat/LongCat-Video",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/05#longcat-video-avatar-1-5"
    },
    {
      "slug": "command-a-plus",
      "date": "2026-05-20",
      "model": "Command A+",
      "lab": "Cohere",
      "lab_slug": "cohere",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "MoE: 218B total, 25B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "API",
        "Open weights",
        "Enterprise"
      ],
      "summary": "Final model in the Command A family and Cohere's first mixture-of-experts model, combining vision input, reasoning, translation and agentic tasks in one model.",
      "highlights": [
        "Supports 48 languages, including all official EU languages",
        "Deployable on one B200 or two H100 GPUs",
        "Supports up to 64K output tokens"
      ],
      "source_title": "Announcing Cohere's Command A+",
      "source_url": "https://docs.cohere.com/v2/changelog/command-a-plus-05-2026",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/05#command-a-plus"
    },
    {
      "slug": "qwen3-7-max",
      "date": "2026-05-20",
      "model": "Qwen3.7-Max",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Qwen3.7-Max is Qwen's proprietary model designed for the agent era, targeting coding, office workflow automation, and long autonomous runs. Qwen demonstrated a 35-hour autonomous kernel optimization run with more than 1,000 tool calls.",
      "highlights": [
        "Built to work across scaffolds such as Claude Code, OpenClaw, and Qwen Code",
        "Supports MCP integrations and multi-agent orchestration",
        "Available through Alibaba Cloud Model Studio"
      ],
      "source_title": "Qwen3.7: The Agent Frontier",
      "source_url": "https://qwen.ai/blog?id=qwen3.7",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/05#qwen3-7-max"
    },
    {
      "slug": "stable-audio-3-large",
      "date": "2026-05-20",
      "model": "Stable Audio 3.0 Large",
      "lab": "Stability AI",
      "lab_slug": "stability-ai",
      "type": "music",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Enterprise"
      ],
      "summary": "Largest Stable Audio 3.0 model with the family's most advanced musicality, built for music platforms and creative applications that need low-latency, high-volume generation.",
      "highlights": [
        "Generates tracks of more than six minutes",
        "Supports single and multi-segment audio inpainting and continuation",
        "Available via the Stability AI API and enterprise self-hosting"
      ],
      "source_title": "Meet Stable Audio 3.0, the model family built for artistic experimentation with open-weight models",
      "source_url": "https://stability.ai/news-updates/meet-stable-audio-3-the-model-family-built-for-artistic-experimentation-with-open-weight-models",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/05#stable-audio-3-large"
    },
    {
      "slug": "stable-audio-3-open",
      "date": "2026-05-20",
      "model": "Stable Audio 3.0 Small, Small SFX and Medium",
      "lab": "Stability AI",
      "lab_slug": "stability-ai",
      "type": "music",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Stability AI Community License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Open-weight members of the Stable Audio 3.0 family, trained on fully licensed data with a semantic-acoustic autoencoder; Small runs full music composition on-device and Medium generates tracks over six minutes.",
      "highlights": [
        "3.0 Small generates up to two minutes; Medium more than six minutes",
        "Small SFX targets on-device sound effects",
        "LoRA training documentation published with the weights"
      ],
      "source_title": "Meet Stable Audio 3.0, the model family built for artistic experimentation with open-weight models",
      "source_url": "https://stability.ai/news-updates/meet-stable-audio-3-the-model-family-built-for-artistic-experimentation-with-open-weight-models",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/05#stable-audio-3-open"
    },
    {
      "slug": "gemini-3-5-flash",
      "date": "2026-05-19",
      "model": "Gemini 3.5 Flash",
      "lab": "Google",
      "lab_slug": "google",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1048576,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms",
        "Consumer app"
      ],
      "summary": "Gemini 3.5 Flash was the first model in the Gemini 3.5 family, which Google built to combine frontier intelligence with action, targeting agents, coding and complex long-horizon tasks at Flash speed.",
      "highlights": [
        "Launched at Google I/O 2026 in the Gemini app, AI Mode in Search, Google Antigravity and the Gemini API",
        "Released directly as the generally available gemini-3.5-flash",
        "Computer use support added in public preview on June 24, 2026",
        "Replaced by Gemini 3.6 Flash, with gemini-3.5-flash requests routed to 3.6 Flash from Oct 8, 2026"
      ],
      "source_title": "Gemini 3.5: frontier intelligence with action",
      "source_url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-5/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/05#gemini-3-5-flash"
    },
    {
      "slug": "gemini-omni-flash",
      "date": "2026-05-19",
      "model": "Gemini Omni Flash",
      "lab": "Google",
      "lab_slug": "google",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video",
        "audio"
      ],
      "output_modalities": [
        "video",
        "audio",
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms",
        "Consumer app"
      ],
      "summary": "Gemini Omni Flash is the first model in Google's Omni family, which combines Gemini's reasoning with generation, starting with video it can create and edit conversationally from text, image, video and audio inputs.",
      "highlights": [
        "Rolled out first to the Gemini app, Google Flow and YouTube Shorts at Google I/O 2026",
        "Public preview in the Gemini API and Google AI Studio (gemini-omni-flash-preview) began June 30, 2026",
        "Only voice references were supported for audio input at launch"
      ],
      "source_title": "Introducing Gemini Omni",
      "source_url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-omni/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/05#gemini-omni-flash"
    },
    {
      "slug": "recraft-v4-1",
      "date": "2026-05-14",
      "model": "Recraft V4.1",
      "lab": "Recraft",
      "lab_slug": "recraft",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Update to Recraft V4 with more natural photorealism, new illustration styles and better aesthetic results from short prompts, plus a Utility variant for simpler outputs.",
      "highlights": [
        "New Utility variant",
        "V4.1 Vector for illustrations and text",
        "Pro variant also offered"
      ],
      "source_title": "Recraft V4.1: More Beautiful by Nature",
      "source_url": "https://www.recraft.ai/blog/recraft-v4-1-more-beautiful-by-nature",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/05#recraft-v4-1"
    },
    {
      "slug": "ernie-5-1",
      "date": "2026-05-09",
      "model": "ERNIE 5.1",
      "lab": "Baidu",
      "lab_slug": "baidu",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "ERNIE 5.1 is Baidu's successor to ERNIE 5.0, derived through an elastic training framework that compresses total parameters to about one-third and active parameters to about one-half of ERNIE 5.0.",
      "highlights": [
        "Four-stage post-training that includes multi-teacher on-policy distillation and online reinforcement learning",
        "Available on the ERNIE website and in an ERNIE 5.1 playground on Baidu AI Studio"
      ],
      "source_title": "ERNIE 5.1 Officially Released! Topping Multiple Leaderboards",
      "source_url": "https://ernie.baidu.com/research/posts/ernie-5.1-0508-release/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/05#ernie-5-1"
    },
    {
      "slug": "emo",
      "date": "2026-05-08",
      "model": "EMO",
      "lab": "Ai2",
      "lab_slug": "ai2",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "MoE: 14B total, 1B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Research mixture-of-experts model pretrained so that modular structure emerges from data, letting a task use about 12.5% of its experts with near full performance.",
      "highlights": [
        "128 experts in total with 8 active",
        "Trained on 1 trillion tokens",
        "Released under Apache 2.0"
      ],
      "source_title": "EMO: Pretraining mixture of experts for emergent modularity | Ai2",
      "source_url": "https://allenai.org/blog/emo",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/05#emo"
    },
    {
      "slug": "stepaudio-2-5-realtime",
      "date": "2026-05-08",
      "model": "StepAudio 2.5 Realtime",
      "lab": "StepFun",
      "lab_slug": "stepfun",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "StepAudio 2.5 Realtime is StepFun's real-time voice conversation model, focused on understanding and expressing paralinguistic cues, user-defined personas set through the API, and stable role-play behavior.",
      "highlights": [
        "Interprets tone, pace, pauses, sighs, and laughter in user speech",
        "Personality, background, preferences, and conversational boundaries can be set through the API",
        "Trained on more than 10,000 native personas expanded into a million-scale persona feature set"
      ],
      "source_title": "StepAudio 2.5 Realtime 上线！真人级实时语音对话能力，千万人设任你打造",
      "source_url": "https://www.stepfun.com/news/stepaudio-2-5-realtime",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/05#stepaudio-2-5-realtime"
    },
    {
      "slug": "gpt-realtime-2",
      "date": "2026-05-07",
      "model": "GPT-Realtime-2",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "audio",
        "image"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": 128000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "GPT-Realtime-2 is a speech-to-speech model for the Realtime API with GPT-5-class reasoning and adjustable reasoning effort, aimed at voice agents that call tools, handle interruptions, and sustain longer sessions.",
      "highlights": [
        "Context window raised from 32K to 128K tokens compared with GPT-Realtime-1.5",
        "Reasoning effort levels minimal, low, medium, high, and xhigh, with low as the default",
        "Supports preambles and parallel tool calls with audible tool transparency",
        "API model ID gpt-realtime-2; testable in the Playground"
      ],
      "source_title": "Advancing voice intelligence with new models in the API | OpenAI",
      "source_url": "https://openai.com/index/advancing-voice-intelligence-with-new-models-in-the-api/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/05#gpt-realtime-2"
    },
    {
      "slug": "gpt-realtime-translate",
      "date": "2026-05-07",
      "model": "GPT-Realtime-Translate",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "audio",
        "text"
      ],
      "context_window": 16000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "GPT-Realtime-Translate is a streaming speech-to-speech translation model that translates live speech from more than 70 input languages into 13 output languages, returning translated audio and transcripts while the speaker is still talking.",
      "highlights": [
        "API model ID gpt-realtime-translate on a dedicated realtime translation endpoint (v1/realtime/translations)",
        "16,000-token context window per the API model page",
        "Launched alongside GPT-Realtime-2 and GPT-Realtime-Whisper"
      ],
      "source_title": "Advancing voice intelligence with new models in the API | OpenAI",
      "source_url": "https://openai.com/index/advancing-voice-intelligence-with-new-models-in-the-api/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/05#gpt-realtime-translate"
    },
    {
      "slug": "gpt-realtime-whisper",
      "date": "2026-05-07",
      "model": "GPT-Realtime-Whisper",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio",
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 16000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "GPT-Realtime-Whisper is a streaming speech-to-text model for the Realtime API that transcribes audio as people speak, built for live captions, meeting notes, and voice agents that need low-latency transcripts.",
      "highlights": [
        "API model ID gpt-realtime-whisper on the Realtime transcription sessions endpoint",
        "Lets developers tune the tradeoff between latency and accuracy",
        "16,000-token context window per the API model page"
      ],
      "source_title": "Advancing voice intelligence with new models in the API | OpenAI",
      "source_url": "https://openai.com/index/advancing-voice-intelligence-with-new-models-in-the-api/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/05#gpt-realtime-whisper"
    },
    {
      "slug": "grok-imagine-image-quality",
      "date": "2026-05-06",
      "model": "Grok Imagine Quality Mode",
      "lab": "xAI",
      "lab_slug": "xai",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Grok Imagine Quality Mode is an image generation and editing model on the Grok Imagine API, served as grok-imagine-image-quality, with higher realism, cleaner multilingual text rendering, and tighter prompt following.",
      "highlights": [
        "API model ID: grok-imagine-image-quality",
        "Launched for enterprise developers and teams via the Grok Imagine API",
        "grok-imagine-image-pro requests redirected to it from May 15, 2026",
        "Retired November 2, 2026; requests are now served by grok-imagine-image-2.0"
      ],
      "source_title": "Grok Imagine Quality Mode API | SpaceXSI",
      "source_url": "https://x.ai/news/grok-imagine-quality-mode",
      "retired_date": "2026-11-02",
      "retirement_note": "xAI retired the grok-imagine-image-quality model on November 2, 2026, after a 60-day notice that began on September 2, and began serving its requests with grok-imagine-image-2.0 at low quality.",
      "retirement_source_url": "https://docs.x.ai/developers/migration/imagine-image-quality-nov-2",
      "permalink": "https://theaiengineering.com/releases/2026/05#grok-imagine-image-quality"
    },
    {
      "slug": "gpt-5-5-instant",
      "date": "2026-05-05",
      "model": "GPT-5.5 Instant",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "GPT-5.5 Instant replaced GPT-5.3 Instant as ChatGPT's default model for all users, with improved factuality, clearer and more concise answers, better analysis of image uploads, and more personalized responses.",
      "highlights": [
        "Available in the API as chat-latest",
        "GPT-5.3 Instant stayed available to paid users for three months before retirement",
        "Updated on May 28, 2026 to improve response style and quality"
      ],
      "source_title": "GPT-5.5 Instant: smarter, clearer, and more personalized | OpenAI",
      "source_url": "https://openai.com/index/gpt-5-5-instant/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/05#gpt-5-5-instant"
    },
    {
      "slug": "granite-switch-4-1",
      "date": "2026-05-05",
      "model": "Granite Switch 4.1",
      "lab": "IBM",
      "lab_slug": "ibm",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "3B, 8B and 30B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Modular models that compose a Granite 4.1 base model with adapters activated by control tokens, so one endpoint can switch between chat, decisions and guardrail checks.",
      "highlights": [
        "Released first as 3B, 8B and 30B previews",
        "Non-preview 8B and 30B versions released October 14, 2026",
        "Revision 4.1.1 on October 27, 2026 added embedded speech-to-text"
      ],
      "source_title": "ibm-granite/granite-switch-4.1-8b-preview · Hugging Face",
      "source_url": "https://huggingface.co/ibm-granite/granite-switch-4.1-8b-preview",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/05#granite-switch-4-1"
    },
    {
      "slug": "molmoact-2",
      "date": "2026-05-05",
      "model": "MolmoAct 2",
      "lab": "Ai2",
      "lab_slug": "ai2",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "action"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Upgrade to Ai2's Action Reasoning Model that pairs the Molmo 2-ER embodied reasoning model with a flow-matching action expert for real-world robots.",
      "highlights": [
        "Molmo 2-ER released as the reasoning backbone",
        "Includes an open action tokenizer",
        "Full code and training data released later"
      ],
      "source_title": "MolmoAct 2: An open foundation for robots that work in the real world | Ai2",
      "source_url": "https://allenai.org/blog/molmoact2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/05#molmoact-2"
    },
    {
      "slug": "luma-uni-1-1",
      "date": "2026-05-05",
      "model": "Uni-1.1",
      "lab": "Luma AI",
      "lab_slug": "luma-ai",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Luma's Unified Intelligence image model, which runs reasoning and image generation in the same model, opened to developers as the Uni-1.1 REST API for generation and natural-language editing.",
      "highlights": [
        "Up to nine reference images per request",
        "Generate Image and Modify Image endpoints",
        "Builds on Uni-1, which Luma launched the previous month"
      ],
      "source_title": "Introducing the Uni-1.1 API: Intelligence You Can Direct. Aesthetic You Can Ship.",
      "source_url": "https://lumalabs.ai/news/uni-1-1-api",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/05#luma-uni-1-1"
    },
    {
      "slug": "granite-4-1",
      "date": "2026-04-29",
      "model": "Granite 4.1",
      "lab": "IBM",
      "lab_slug": "ibm",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 131072,
      "parameters": "3B, 8B and 30B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Dense Granite 4.1 language models with an improved post-training pipeline, released as base and instruct checkpoints in 3B, 8B and 30B sizes.",
      "highlights": [
        "Supports 12 languages including Arabic, Japanese and Chinese",
        "Granite Guardian 4.1 8B, fine-tuned from Granite 4.1 8B, was also released in April 2026",
        "Released under Apache 2.0"
      ],
      "source_title": "ibm-granite/granite-4.1-8b · Hugging Face",
      "source_url": "https://huggingface.co/ibm-granite/granite-4.1-8b",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#granite-4-1"
    },
    {
      "slug": "granite-embedding-multilingual-r2",
      "date": "2026-04-29",
      "model": "Granite Embedding Multilingual R2",
      "lab": "IBM",
      "lab_slug": "ibm",
      "type": "embedding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "embedding"
      ],
      "context_window": 32768,
      "parameters": "311M and 97M",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Multilingual Granite embedding models rebuilt on ModernBERT, supporting 200+ languages and code retrieval with a context window extended from 512 to 32,768 tokens.",
      "highlights": [
        "311M model outputs 768-dimensional vectors",
        "Matryoshka truncation down to 128 dimensions",
        "Released under Apache 2.0"
      ],
      "source_title": "ibm-granite/granite-embedding-311m-multilingual-r2 · Hugging Face",
      "source_url": "https://huggingface.co/ibm-granite/granite-embedding-311m-multilingual-r2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#granite-embedding-multilingual-r2"
    },
    {
      "slug": "granite-speech-4-1",
      "date": "2026-04-29",
      "model": "Granite Speech 4.1",
      "lab": "IBM",
      "lab_slug": "ibm",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "2B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Speech-language model for multilingual speech recognition and translation in English, French, German, Spanish, Portuguese and Japanese, with punctuation and truecasing.",
      "highlights": [
        "Plus variant adds speaker-attributed ASR and word timestamps",
        "NAR variant uses a non-autoregressive architecture",
        "Released under Apache 2.0"
      ],
      "source_title": "ibm-granite/granite-speech-4.1-2b · Hugging Face",
      "source_url": "https://huggingface.co/ibm-granite/granite-speech-4.1-2b",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#granite-speech-4-1"
    },
    {
      "slug": "granite-vision-4-1",
      "date": "2026-04-29",
      "model": "Granite Vision 4.1",
      "lab": "IBM",
      "lab_slug": "ibm",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "4B: 3.4B LLM + 0.6B vision",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Vision-language model built on Granite 4.1 3B for structured document extraction, covering chart conversion, complex table extraction and semantic key-value pair extraction.",
      "highlights": [
        "Chart extraction to CSV, summaries and code",
        "Trained with the ChartNet dataset",
        "Released under Apache 2.0"
      ],
      "source_title": "ibm-granite/granite-vision-4.1-4b · Hugging Face",
      "source_url": "https://huggingface.co/ibm-granite/granite-vision-4.1-4b",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#granite-vision-4-1"
    },
    {
      "slug": "step-image-edit-2",
      "date": "2026-04-29",
      "model": "Step Image Edit 2",
      "lab": "StepFun",
      "lab_slug": "stepfun",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "3.5B",
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Step Image Edit 2 is a lightweight 3.5B-parameter StepFun model for image generation and editing, supporting Chinese and English text rendering, local edits, subject consistency, and style transfer with fast response times.",
      "highlights": [
        "Generates or edits an image in about 0.5 to 2 seconds",
        "Launched on the StepFun open platform and Step Plan",
        "StepFun retired the model and its image APIs on October 10, 2026"
      ],
      "source_title": "阶跃 Step Image Edit 2 上线！小身材、大能力、快响应",
      "source_url": "https://www.stepfun.com/news/step-image-edit-2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#step-image-edit-2"
    },
    {
      "slug": "mistral-medium-3-5",
      "date": "2026-04-28",
      "model": "Mistral Medium 3.5",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "128B",
      "model_license": "Modified MIT License",
      "available_at_launch": [
        "Open weights",
        "API",
        "Consumer app"
      ],
      "summary": "Dense 128B flagship that merges instruction following, reasoning and coding in one set of weights, with per-request reasoning effort and a vision encoder trained from scratch.",
      "highlights": [
        "Showcased May 22, 2026 as the default model for remote agents in Vibe and Le Chat",
        "Self-hostable on as few as four GPUs",
        "API model ID mistral-medium-3-5"
      ],
      "source_title": "Mistral Medium 3.5 | Mistral Docs",
      "source_url": "https://docs.mistral.ai/models/mistral-medium-3-5-26-04",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#mistral-medium-3-5"
    },
    {
      "slug": "nemotron-3-nano-omni",
      "date": "2026-04-28",
      "model": "Nemotron 3 Nano Omni",
      "lab": "NVIDIA",
      "lab_slug": "nvidia",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video",
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "MoE: 31B total, about 3B active",
      "model_license": "NVIDIA Open Model Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Omnimodal reasoning model built on the Nemotron 3 Nano 30B hybrid Mamba-Transformer backbone that understands video, audio, images and text for Q&A, transcription and document intelligence.",
      "highlights": [
        "Maximum context of 256K tokens",
        "Activates about 3B parameters per token",
        "Released under the NVIDIA Open Model Agreement"
      ],
      "source_title": "nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16 · Hugging Face",
      "source_url": "https://huggingface.co/nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#nemotron-3-nano-omni"
    },
    {
      "slug": "deepseek-v4",
      "date": "2026-04-24",
      "model": "DeepSeek-V4 (Preview)",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": "MoE: 1.6T total, 49B active (V4-Pro); 284B total, 13B active (V4-Flash)",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "DeepSeek-V4 is a preview release of two open-weights mixture-of-experts models, V4-Pro and V4-Flash. It introduced a hybrid attention design combining token-wise compression with DeepSeek Sparse Attention and made 1M-token context the default across DeepSeek services.",
      "highlights": [
        "Both models pre-trained on more than 32T tokens and support thinking and non-thinking modes",
        "API supports OpenAI ChatCompletions and Anthropic formats via deepseek-v4-pro and deepseek-v4-flash",
        "Legacy deepseek-chat and deepseek-reasoner names were scheduled for retirement on July 24, 2026"
      ],
      "source_title": "DeepSeek V4 Preview Release",
      "source_url": "https://api-docs.deepseek.com/news/news260424",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#deepseek-v4"
    },
    {
      "slug": "stepaudio-2-5-asr",
      "date": "2026-04-24",
      "model": "StepAudio 2.5 ASR",
      "lab": "StepFun",
      "lab_slug": "stepfun",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "StepAudio 2.5 ASR is StepFun's speech recognition model that brings Step 3.5 Flash's multi-token prediction to transcription and reuses a 32K language model context to transcribe up to 30 minutes of audio in one pass.",
      "highlights": [
        "ASR plus MTP-5 architecture with peak inference of 500 tokens per second",
        "Transcribes up to 30 minutes of audio in a single request without slicing",
        "Launched on the StepFun open platform and Step Plan"
      ],
      "source_title": "阶跃 StepAudio 2.5 ASR 上线！500TPS 极速推理，30分钟语音“秒级转写”",
      "source_url": "https://www.stepfun.com/news/stepaudio-2-5-asr",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#stepaudio-2-5-asr"
    },
    {
      "slug": "gpt-5-5",
      "date": "2026-04-23",
      "model": "GPT-5.5 (with GPT-5.5 Pro)",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1050000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API",
        "Enterprise"
      ],
      "summary": "GPT-5.5 is a frontier model for complex professional work such as coding, research, data analysis, and operating software, which OpenAI says matches GPT-5.4 per-token latency while using fewer tokens on Codex tasks.",
      "highlights": [
        "GPT-5.5 Thinking for Plus, Pro, Business, and Enterprise; GPT-5.5 Pro for Pro, Business, and Enterprise",
        "API models gpt-5.5 and gpt-5.5-pro became available on April 24, 2026",
        "1,050,000-token context window per the API model page; 400K context in Codex",
        "Reasoning effort defaults to medium in the API"
      ],
      "source_title": "Introducing GPT-5.5 | OpenAI",
      "source_url": "https://openai.com/index/introducing-gpt-5-5/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#gpt-5-5"
    },
    {
      "slug": "grok-voice-1-0",
      "date": "2026-04-23",
      "model": "Grok Voice 1.0",
      "lab": "xAI",
      "lab_slug": "xai",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Grok Voice 1.0 is SpaceXSI's flagship voice model for the Speech to Speech API, built for complex, multi-step voice agent workflows in customer support and sales that need precise data entry and heavy tool calling.",
      "highlights": [
        "Available via the Speech to Speech (voice agent) API",
        "Natively supports 25+ languages",
        "Built to reason through edge cases before responding",
        "Developed with partners including Starlink for noisy telephony conditions"
      ],
      "source_title": "Grok Voice 1.0 | SpaceXSI",
      "source_url": "https://x.ai/news/grok-voice-think-fast-1",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#grok-voice-1-0"
    },
    {
      "slug": "hy3-preview",
      "date": "2026-04-23",
      "model": "Hy3 preview",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "MoE: 295B total, 21B active",
      "model_license": "Tencent HY Community License Agreement",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Hy3 preview is the first model trained on Tencent's rebuilt Hy infrastructure, an open mixture-of-experts model with three reasoning modes that improves complex reasoning, instruction following, context learning, coding, and agent tasks.",
      "highlights": [
        "295B total parameters with 21B active and a 256K context window",
        "Weights on GitHub, Hugging Face, ModelScope, and GitCode with vLLM and SGLang support",
        "API on Tencent Cloud, plus Token Plans for individuals"
      ],
      "source_title": "Hy3 preview: The First Step in Rebuilding the Hy model",
      "source_url": "https://hy.tencent.com/research/hy3-preview",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#hy3-preview"
    },
    {
      "slug": "mimo-v2-5-asr",
      "date": "2026-04-23",
      "model": "MiMo-V2.5-ASR",
      "lab": "Xiaomi",
      "lab_slug": "xiaomi",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "MiMo-V2.5-ASR is Xiaomi's open-source speech recognition model for bilingual Chinese and English, Chinese dialects, code-switching, lyrics transcription, knowledge-heavy content, and noisy or multi-speaker audio.",
      "highlights": [
        "Open-sourced with the MiMo-V2.5-TTS series launch on April 23, 2026",
        "Supports Wu, Cantonese, Hokkien, Sichuanese, and other dialects",
        "Added to the MiMo API Platform on June 2, 2026",
        "Weights under the MIT license"
      ],
      "source_title": "MiMo-V2.5-ASR | Xiaomi",
      "source_url": "https://mimo.xiaomi.com/mimo-v2-5-asr/index.html",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#mimo-v2-5-asr"
    },
    {
      "slug": "mimo-v2-5-pro",
      "date": "2026-04-23",
      "model": "MiMo-V2.5-Pro",
      "lab": "Xiaomi",
      "lab_slug": "xiaomi",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": "MoE: 1.02T total, 42B active",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "MiMo-V2.5-Pro is Xiaomi's flagship agentic language model, improving on MiMo-V2-Pro in general agent tasks, complex software engineering, and long-horizon work spanning more than a thousand tool calls.",
      "highlights": [
        "Public beta on the MiMo API Platform began April 23, 2026",
        "Weights open-sourced under the MIT license on April 27, 2026",
        "Hybrid attention with a 6:1 sliding-window to global ratio and three MTP modules"
      ],
      "source_title": "MiMo-V2.5-Pro | Xiaomi",
      "source_url": "https://mimo.xiaomi.com/mimo-v2-5-pro/index.html",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#mimo-v2-5-pro"
    },
    {
      "slug": "mimo-v2-5-tts",
      "date": "2026-04-23",
      "model": "MiMo-V2.5-TTS Series",
      "lab": "Xiaomi",
      "lab_slug": "xiaomi",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "MiMo-V2.5-TTS is a family of three Xiaomi speech synthesis models covering premium stock voices, voice design from a text description, and voice cloning, all steerable through style instructions and inline audio tags.",
      "highlights": [
        "Models: mimo-v2.5-tts, mimo-v2.5-tts-voicedesign, and mimo-v2.5-tts-voiceclone",
        "Released on the MiMo API Platform on April 23, 2026, per its model release log",
        "Accepts director-script style input describing characters, scenes, and per-line instructions"
      ],
      "source_title": "MiMo-V2.5-TTS Series | Xiaomi",
      "source_url": "https://mimo.xiaomi.com/mimo-v2-5-tts/index.html",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#mimo-v2-5-tts"
    },
    {
      "slug": "seed3d-2-0",
      "date": "2026-04-23",
      "model": "Seed3D 2.0",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "image"
      ],
      "output_modalities": [
        "3d"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Seed3D 2.0 is ByteDance Seed's second-generation image-to-3D model, with an upgraded architecture for higher geometric precision and material quality so generated assets come closer to production use in embodied AI and manufacturing.",
      "highlights": [
        "Uses a larger DiT for coarse geometry generation",
        "Adopts a mixture-of-experts design for high-resolution material generation",
        "API available on Volcano Engine Ark as Doubao-Seed3D-2.0",
        "Technical report published alongside the release"
      ],
      "source_title": "Seed3D 2.0 Released: Higher Precision and Greater Usability",
      "source_url": "https://seed.bytedance.com/en/blog/seed3d-2-0-released-higher-precision-and-greater-usability",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#seed3d-2-0"
    },
    {
      "slug": "mimo-v2-5",
      "date": "2026-04-22",
      "model": "MiMo-V2.5",
      "lab": "Xiaomi",
      "lab_slug": "xiaomi",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video",
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": "MoE: 310B total, 15B active",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "MiMo-V2.5 is a natively omnimodal Xiaomi model that adds in-house visual and audio encoders to the MiMo-V2-Flash backbone, built for agentic tasks with up to 1M tokens of context.",
      "highlights": [
        "Trained on about 48T tokens",
        "Xiaomi reports agentic performance above MiMo-V2-Pro",
        "Weights, including the Base model, released under MIT on April 27, 2026",
        "Public beta on the MiMo API Platform from April 23, 2026"
      ],
      "source_title": "MiMo-V2.5 | Xiaomi",
      "source_url": "https://mimo.xiaomi.com/mimo-v2-5/index.html",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#mimo-v2-5"
    },
    {
      "slug": "openai-privacy-filter",
      "date": "2026-04-22",
      "model": "OpenAI Privacy Filter",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "1.5B total, 50M active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "OpenAI Privacy Filter is an open-weight model for detecting and redacting personally identifiable information in text, built as a bidirectional token classifier that labels input in a single pass and can run locally.",
      "highlights": [
        "1.5B total parameters with 50M active parameters",
        "Supports up to 128,000 tokens of context",
        "Released under Apache 2.0 on Hugging Face (openai/privacy-filter) and GitHub",
        "Predicts PII spans across eight privacy categories"
      ],
      "source_title": "Introducing OpenAI Privacy Filter | OpenAI",
      "source_url": "https://openai.com/index/introducing-openai-privacy-filter/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#openai-privacy-filter"
    },
    {
      "slug": "qwen3-6-27b",
      "date": "2026-04-22",
      "model": "Qwen3.6-27B",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 262144,
      "parameters": "27B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Consumer app",
        "Open weights"
      ],
      "summary": "Qwen3.6-27B is a dense 27B open-weights multimodal model focused on agentic coding. Qwen says it surpasses the previous open flagship, Qwen3.5-397B-A17B, on major coding benchmarks while being simpler to deploy.",
      "highlights": [
        "Supports multimodal thinking and non-thinking modes",
        "Native 262,144-token context, extensible to 1,010,000 tokens",
        "Available on Qwen Studio; weights under the Apache 2.0 license"
      ],
      "source_title": "Qwen3.6-27B: Flagship-Level Coding in a 27B Dense Model",
      "source_url": "https://qwen.ai/blog?id=qwen3.6-27b",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#qwen3-6-27b"
    },
    {
      "slug": "gpt-image-2",
      "date": "2026-04-21",
      "model": "GPT Image 2 (ChatGPT Images 2.0)",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "ChatGPT Images 2.0 introduced GPT Image 2, OpenAI's new image generation and editing model for ChatGPT and the API, adding flexible image sizes, high-fidelity image inputs, and images with thinking for paid ChatGPT plans.",
      "highlights": [
        "Available on all ChatGPT plans; images with thinking on paid plans with Thinking and Pro models",
        "API model ID gpt-image-2 (snapshot gpt-image-2-2026-04-21) with Batch API support",
        "Transparent backgrounds added in preview for gpt-image-2 on August 20, 2026"
      ],
      "source_title": "Introducing ChatGPT Images 2.0 | OpenAI",
      "source_url": "https://openai.com/index/introducing-chatgpt-images-2-0/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#gpt-image-2"
    },
    {
      "slug": "kimi-k2-6",
      "date": "2026-04-20",
      "model": "Kimi K2.6",
      "lab": "Moonshot AI",
      "lab_slug": "moonshot-ai",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "MoE: 1T total, 32B active",
      "model_license": "Modified MIT License",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "Kimi K2.6 is Moonshot AI's open-weights model focused on long-horizon coding, long-running execution, and agent swarm capabilities. It improves on Kimi K2.5 across programming languages and tasks such as front-end, DevOps, and performance optimization.",
      "highlights": [
        "Available on Kimi.com, the Kimi App, the API, and Kimi Code",
        "Native multimodal model with the 400M-parameter MoonViT vision encoder",
        "Weights and code under the Modified MIT License"
      ],
      "source_title": "Kimi K2.6: Advancing Open-Source Coding",
      "source_url": "https://www.kimi.com/en/blog/kimi-k2-6",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#kimi-k2-6"
    },
    {
      "slug": "qwen3-6-max-preview",
      "date": "2026-04-18",
      "model": "Qwen3.6-Max-Preview",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Qwen3.6-Max-Preview is an early preview of Qwen's next proprietary flagship, released after Qwen3.6-Plus. It brings stronger world knowledge, instruction following, and agentic coding than Qwen3.6-Plus.",
      "highlights": [
        "Available on Qwen Studio",
        "API model name qwen3.6-max-preview on Alibaba Cloud Model Studio",
        "Qwen describes it as still under active development"
      ],
      "source_title": "Qwen3.6-Max-Preview: Smarter, Sharper, Still Evolving",
      "source_url": "https://qwen.ai/blog?id=qwen3.6-max-preview",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#qwen3-6-max-preview"
    },
    {
      "slug": "grok-speech-to-text",
      "date": "2026-04-17",
      "model": "Grok Speech to Text",
      "lab": "xAI",
      "lab_slug": "xai",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Grok Speech to Text is a standalone transcription API built on the stack behind Grok Voice, offering batch REST and real-time WebSocket modes with word-level timestamps, speaker diarization, and multichannel support.",
      "highlights": [
        "Supports 25+ languages",
        "Inverse Text Normalization formats numbers, dates, and currencies",
        "Docs release notes list general availability on April 15, 2026",
        "Docs later refer to it as grok-voice-transcribe-1.0"
      ],
      "source_title": "Grok Speech to Text and Text to Speech APIs | SpaceXSI",
      "source_url": "https://x.ai/news/grok-stt-and-tts-apis",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#grok-speech-to-text"
    },
    {
      "slug": "grok-text-to-speech",
      "date": "2026-04-17",
      "model": "Grok Text to Speech",
      "lab": "xAI",
      "lab_slug": "xai",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Grok Text to Speech is a standalone speech synthesis API built on the stack behind Grok Voice, with REST and real-time WebSocket modes and inline speech tags for laughter, whispers, emphasis, and pauses.",
      "highlights": [
        "Docs release notes list general availability on March 16, 2026",
        "Announced alongside Grok Speech to Text",
        "All voices can speak every supported language; Eve is the default voice",
        "Output formats range from MP3 to telephony-optimized mu-law"
      ],
      "source_title": "Grok Speech to Text and Text to Speech APIs | SpaceXSI",
      "source_url": "https://x.ai/news/grok-stt-and-tts-apis",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#grok-text-to-speech"
    },
    {
      "slug": "claude-opus-4-7",
      "date": "2026-04-16",
      "model": "Claude Opus 4.7",
      "lab": "Anthropic",
      "lab_slug": "anthropic",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Cloud platforms"
      ],
      "summary": "Claude Opus 4.7 is an upgrade to Claude Opus 4.6 focused on advanced software engineering, higher-resolution vision, and instruction following. It is the first model Anthropic shipped with new safeguards that automatically detect and block prohibited or high-risk cybersecurity uses.",
      "highlights": [
        "API model ID claude-opus-4-7; 1M-token context window and 128K max output tokens",
        "Updated tokenizer can map the same input to roughly 1.0 to 1.35 times as many tokens",
        "Maximum image resolution raised to 2,576 pixels on the long edge",
        "Adds an xhigh effort level and task budgets in beta"
      ],
      "source_title": "Introducing Claude Opus 4.7",
      "source_url": "https://www.anthropic.com/news/claude-opus-4-7",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#claude-opus-4-7"
    },
    {
      "slug": "gpt-rosalind",
      "date": "2026-04-16",
      "model": "GPT-Rosalind",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [],
      "output_modalities": [],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Enterprise"
      ],
      "summary": "GPT-Rosalind is OpenAI's frontier reasoning model for biology, drug discovery, and translational medicine, and the first model in its GPT-Rosalind life sciences series. It launched through a trusted access program for qualified Enterprise customers in the U.S.",
      "highlights": [
        "Available in ChatGPT, Codex, and the API for qualified customers through trusted access",
        "Launched alongside a free Life Sciences research plugin for Codex that connects to more than 50 scientific tools and data sources",
        "Launch customers included Amgen, Moderna, the Allen Institute, and Thermo Fisher Scientific",
        "Left research preview on September 11, 2026"
      ],
      "source_title": "Introducing GPT-Rosalind for life sciences research | OpenAI",
      "source_url": "https://openai.com/index/introducing-gpt-rosalind/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#gpt-rosalind"
    },
    {
      "slug": "hy-world-2-0",
      "date": "2026-04-16",
      "model": "HY-World 2.0",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "3d"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Tencent HY-World 2.0 Community License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "HY-World 2.0 is an open multi-modal world model framework that generates and reconstructs navigable 3D worlds from text, single images, multi-view images, or videos, producing meshes or Gaussian splatting representations.",
      "highlights": [
        "WorldMirror 2.0 weights and the technical report released April 16, 2026",
        "HY-Pano 2.0 and WorldStereo 2.0 weights followed in May 2026",
        "Follows HunyuanWorld 1.0, 1.1 (WorldMirror), and 1.5 (WorldPlay)"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/HY-World-2.0: HY-World 2.0: A Multi-Modal World Model for Reconstructing, Generating, and Simulating 3D Worlds",
      "source_url": "https://github.com/Tencent-Hunyuan/HY-World-2.0",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#hy-world-2-0"
    },
    {
      "slug": "stepaudio-2-5-tts",
      "date": "2026-04-16",
      "model": "StepAudio 2.5 TTS",
      "lab": "StepFun",
      "lab_slug": "stepfun",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "StepAudio 2.5 TTS is StepFun's speech generation model with passage-level and in-text control over emotion, tone, rhythm, and pauses, plus zero-shot voice cloning with adjustable emotion and speaking style.",
      "highlights": [
        "Global context control sets the emotional tone, character state, and scene for a whole passage",
        "Controls can be given as natural-language instructions",
        "Launched on the StepFun open platform and Step Plan"
      ],
      "source_title": "阶跃 StepAudio 2.5 TTS 上线！人人都能是配音导演",
      "source_url": "https://www.stepfun.com/news/stepaudio-2-5-tts",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#stepaudio-2-5-tts"
    },
    {
      "slug": "ernie-image",
      "date": "2026-04-15",
      "model": "ERNIE-Image",
      "lab": "Baidu",
      "lab_slug": "baidu",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "8B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "ERNIE-Image is Baidu's open text-to-image model, built on a single-stream Diffusion Transformer with a lightweight Prompt Enhancer, and released alongside a distilled ERNIE-Image-Turbo variant for faster generation.",
      "highlights": [
        "8B DiT parameters; runs on consumer GPUs with 24GB of VRAM",
        "ERNIE-Image uses about 50 inference steps; ERNIE-Image-Turbo uses 8 steps after DMD and RL optimization",
        "Targets dense text rendering, posters, comics, and multi-panel layouts"
      ],
      "source_title": "baidu/ERNIE-Image · Hugging Face",
      "source_url": "https://huggingface.co/baidu/ERNIE-Image",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#ernie-image"
    },
    {
      "slug": "gemini-3-1-flash-tts",
      "date": "2026-04-15",
      "model": "Gemini 3.1 Flash TTS",
      "lab": "Google",
      "lab_slug": "google",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": 8192,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms"
      ],
      "summary": "Gemini 3.1 Flash TTS is a cost-efficient, expressive text-to-speech model that adds granular audio tags for directing vocal style, pace and delivery with natural language across more than 70 languages.",
      "highlights": [
        "Preview in the Gemini API and Google AI Studio, and for enterprises on Vertex AI",
        "All generated audio is watermarked with SynthID",
        "Streaming output support added on June 17, 2026"
      ],
      "source_title": "Gemini 3.1 Flash TTS: New text-to-speech AI model",
      "source_url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-flash-tts/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#gemini-3-1-flash-tts"
    },
    {
      "slug": "qwen3-6-35b-a3b",
      "date": "2026-04-15",
      "model": "Qwen3.6-35B-A3B",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 262144,
      "parameters": "MoE: 35B total, 3B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "Qwen3.6-35B-A3B is a sparse open-weights mixture-of-experts model focused on agentic coding, surpassing Qwen3.5-35B-A3B by a wide margin. It supports multimodal thinking and non-thinking modes.",
      "highlights": [
        "Offered on the Model Studio API as Qwen3.6-Flash",
        "Native 262,144-token context, extensible to 1,010,000 tokens",
        "Weights on Hugging Face and ModelScope under the Apache 2.0 license"
      ],
      "source_title": "Qwen3.6-35B-A3B: Agentic Coding Power, Now Open to All",
      "source_url": "https://qwen.ai/blog?id=qwen3.6-35b-a3b",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#qwen3-6-35b-a3b"
    },
    {
      "slug": "gemini-robotics-er-1-6",
      "date": "2026-04-14",
      "model": "Gemini Robotics-ER 1.6",
      "lab": "Google",
      "lab_slug": "google",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 131072,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Gemini Robotics-ER 1.6 is an updated embodied reasoning model for robotics with new capabilities such as instrument reading and improved spatial and physical reasoning, available to developers in preview.",
      "highlights": [
        "Preview model gemini-robotics-er-1.6-preview in the Gemini API and Google AI Studio",
        "Replaced gemini-robotics-er-1.5-preview, which shut down on Apr 30, 2026",
        "Input token limit of 131,072 per the Gemini API model page"
      ],
      "source_title": "Gemini Robotics-ER 1.6",
      "source_url": "https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-robotics-er-1-6/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#gemini-robotics-er-1-6"
    },
    {
      "slug": "mai-image-2-efficient",
      "date": "2026-04-14",
      "model": "MAI-Image-2-Efficient",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "MAI-Image-2-Efficient is a faster, lower-cost version of Microsoft AI's MAI-Image-2 text-to-image model, aimed at high-volume production work such as product shots, marketing creatives and batch pipelines.",
      "highlights": [
        "Available in Microsoft Foundry and MAI Playground at launch",
        "Rolling out across Copilot and Bing",
        "Handles short-form in-image text such as headlines and labels"
      ],
      "source_title": "MAI-Image-2-Efficient: Flagship Quality, 41% Lower Cost",
      "source_url": "https://microsoft.ai/news/mai-image-2-efficient/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#mai-image-2-efficient"
    },
    {
      "slug": "midjourney-v8-1",
      "date": "2026-04-14",
      "model": "Midjourney V8.1",
      "lab": "Midjourney",
      "lab_slug": "midjourney",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Second V8 release with a familiar V7-like aesthetic, a faster HD mode made the default, the return of image prompts and weights, and an updated Describe tool.",
      "highlights": [
        "Became the default model on June 10, 2026",
        "HD images render at 2K without upscaling",
        "Standard jobs render about 4 to 5 times faster than earlier versions"
      ],
      "source_title": "Version - Midjourney",
      "source_url": "https://docs.midjourney.com/hc/en-us/articles/32199405667853-Version",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#midjourney-v8-1"
    },
    {
      "slug": "minimax-music-2-6",
      "date": "2026-04-10",
      "model": "MiniMax Music 2.6",
      "lab": "MiniMax",
      "lab_slug": "minimax",
      "type": "music",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "MiniMax Music 2.6 updates MiniMax's music generation model with a new Cover mode for reworking existing songs, a reworked low end for stronger bass and drums, and faster first output.",
      "highlights": [
        "Cover mode is new in Music 2.6",
        "First-packet latency reduced to under 20 seconds",
        "Supports original, instrumental, and Cover generation modes through the API"
      ],
      "source_title": "MiniMax Music 2.6: Four Stories We Want to Tell",
      "source_url": "https://www.minimax.io/news/music-26",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#minimax-music-2-6"
    },
    {
      "slug": "hy-embodied-0-5",
      "date": "2026-04-09",
      "model": "HY-Embodied-0.5",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "MoT: 4B total, 2.2B active",
      "model_license": "Tencent HY Community License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "HY-Embodied-0.5 is a suite of embodied foundation models for real-world agents, using a Mixture-of-Transformers architecture for spatial-temporal perception and embodied reasoning. Tencent released the edge-oriented MoT-2B variant as open weights.",
      "highlights": [
        "MoT-2B has 4B total parameters with 2.2B activated during inference",
        "A 32B variant is described for complex reasoning; reasoning is distilled into the 2B model",
        "Released with official inference code"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/HY-Embodied: HY-Embodied: Embodied Foundation Models for Real-World Agents",
      "source_url": "https://github.com/Tencent-Hunyuan/HY-Embodied",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#hy-embodied-0-5"
    },
    {
      "slug": "seeduplex",
      "date": "2026-04-09",
      "model": "Seeduplex",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Seeduplex is a native full-duplex speech model that listens while speaking, replacing the earlier half-duplex Doubao end-to-end voice model to improve conversational pacing, interruption handling, and suppression of background interference.",
      "highlights": [
        "Decides on its own whether to reply, keep listening, or yield to an interruption",
        "Can draw on background environmental audio as conversational context",
        "Fully rolled out in the Doubao app's voice call mode"
      ],
      "source_title": "Introducing Seed Full-Duplex Speech LLM: Attentive Listening, Robust Interference Suppression, Enabling More Natural Interaction",
      "source_url": "https://seed.bytedance.com/en/blog/introducing-seed-full-duplex-speech-llm-attentive-listening-robust-interference-suppression-enabling-more-natural-interaction",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#seeduplex"
    },
    {
      "slug": "lfm2-5-vl-450m",
      "date": "2026-04-08",
      "model": "LFM2.5-VL-450M",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32768,
      "parameters": "450M",
      "model_license": "LFM Open License v1.0",
      "available_at_launch": [
        "Consumer app",
        "Open weights"
      ],
      "summary": "An updated 450M vision-language model that adds bounding-box object detection, function calling support, and better instruction following over LFM2-VL-450M, with pretraining scaled from 10T to 28T tokens.",
      "highlights": [
        "Improved multilingual image understanding across eight languages",
        "Built on the LFM2.5-350M backbone per model card",
        "Available on Hugging Face, LEAP, and Liquid Playground"
      ],
      "source_title": "LFM2.5-VL-450M: Structured Visual Intelligence, Edge to Cloud",
      "source_url": "https://www.liquid.ai/blog/lfm2-5-vl-450m",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#lfm2-5-vl-450m"
    },
    {
      "slug": "muse-spark",
      "date": "2026-04-08",
      "model": "Muse Spark",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Muse Spark is the first model from Meta Superintelligence Labs, a natively multimodal reasoning model with tool use, visual chain of thought and multi-agent orchestration that powers Meta AI.",
      "highlights": [
        "Available at launch on meta.ai and in the Meta AI app",
        "Contemplating mode runs multiple agents that reason in parallel",
        "API access opened only as a private preview for select users",
        "Weights were not released; Meta said it hopes to open-source future versions"
      ],
      "source_title": "Introducing Muse Spark: Scaling Towards Personal Superintelligence",
      "source_url": "https://ai.meta.com/blog/introducing-muse-spark-msl/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#muse-spark"
    },
    {
      "slug": "glm-5-1",
      "date": "2026-04-07",
      "model": "GLM-5.1",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 200000,
      "parameters": "MoE: 744B total, 40B active",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "GLM-5.1 is Z.ai's flagship model for agentic engineering, designed for long-horizon tasks. Z.ai says it can work independently for up to 8 hours in a single run, from planning and execution to delivery.",
      "highlights": [
        "Trained with multi-turn SFT, RL, and a process-quality evaluation framework",
        "Up to 128K output tokens",
        "Weights under the MIT license"
      ],
      "source_title": "GLM-5.1: Towards Long-Horizon Tasks",
      "source_url": "https://z.ai/blog/glm-5.1",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#glm-5-1"
    },
    {
      "slug": "hy-omniweaving",
      "date": "2026-04-03",
      "model": "HY-OmniWeaving",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Tencent HY Community License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "HY-OmniWeaving is an open unified video generation model from the HunyuanVideo team, built on HunyuanVideo 1.5, that supports free-form multimodal composition and reasoning-augmented video generation in a single model.",
      "highlights": [
        "Built on the HunyuanVideo 1.5 backbone",
        "Released with the IntelligentVBench benchmark",
        "Code and weights released April 3, 2026"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/OmniWeaving: Official Implementation of OmniWeaving: Towards Unified Video Generation with Free-form Composition and Reasoning",
      "source_url": "https://github.com/Tencent-Hunyuan/OmniWeaving",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#hy-omniweaving"
    },
    {
      "slug": "gemma-4",
      "date": "2026-04-02",
      "model": "Gemma 4",
      "lab": "Google",
      "lab_slug": "google",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video",
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 262144,
      "parameters": "E2B (2.3B effective), E4B (4.5B effective), 26B MoE (3.8B active), 31B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "API",
        "Cloud platforms"
      ],
      "summary": "Gemma 4 is Google DeepMind's fourth generation of open models, released in E2B, E4B, 26B mixture-of-experts and 31B dense sizes under Apache 2.0. All sizes accept images and video, and the edge models add audio input.",
      "highlights": [
        "Context window of 128K tokens for E2B and E4B and 256K for 26B and 31B",
        "26B MoE has 25.2B total and 3.8B active parameters",
        "Native function calling, structured JSON output and system instructions; trained on 140+ languages",
        "Unquantized 31B fits on a single 80GB NVIDIA H100"
      ],
      "source_title": "Gemma 4: Byte for byte, the most capable open models",
      "source_url": "https://blog.google/innovation-and-ai/technology/developers-tools/gemma-4/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#gemma-4"
    },
    {
      "slug": "glm-5v-turbo",
      "date": "2026-04-02",
      "model": "GLM-5V-Turbo",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video",
        "file"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 200000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "GLM-5V-Turbo is Z.ai's multimodal coding foundation model, combining visual understanding with coding ability. It targets GUI agents and coding agents that need to see an environment, plan actions, and execute tasks.",
      "highlights": [
        "Strengthened fine-grained, geometric, and spatial understanding",
        "Adds multimodal tools such as box drawing, screenshots, and web page reading",
        "Up to 128K output tokens"
      ],
      "source_title": "Model and product release records (BigModel docs)",
      "source_url": "https://docs.bigmodel.cn/cn/update/new-releases",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#glm-5v-turbo"
    },
    {
      "slug": "mai-transcribe-1",
      "date": "2026-04-02",
      "model": "MAI-Transcribe-1",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "MAI-Transcribe-1 is Microsoft AI's first speech-to-text model, a multilingual transcription model supporting 25 languages and built for noisy, real-world audio such as meetings, call centers and voice agents.",
      "highlights": [
        "Supports 25 languages",
        "Available in Microsoft Foundry and MAI Playground",
        "Launched alongside Foundry availability for MAI-Voice-1 and MAI-Image-2"
      ],
      "source_title": "State of the Art Speech Recognition with MAI-Transcribe-1",
      "source_url": "https://microsoft.ai/news/state-of-the-art-speech-recognition-with-mai-transcribe-1/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#mai-transcribe-1"
    },
    {
      "slug": "qwen3-6-plus",
      "date": "2026-04-02",
      "model": "Qwen3.6-Plus",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Qwen3.6-Plus is Qwen's hosted model that succeeded Qwen3.5-Plus, with major upgrades to agentic coding from front-end work to repository-level tasks. It also improves multimodal perception and reasoning.",
      "highlights": [
        "1M-token context window by default",
        "Addresses community feedback from the Qwen3.5-Plus deployment",
        "Available through Alibaba Cloud Model Studio"
      ],
      "source_title": "Qwen3.6-Plus: Towards Real World Agents",
      "source_url": "https://qwen.ai/blog?id=qwen3.6",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/04#qwen3-6-plus"
    },
    {
      "slug": "lfm2-5-350m",
      "date": "2026-03-31",
      "model": "LFM2.5-350M",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32768,
      "parameters": "350M",
      "model_license": "LFM Open License v1.0",
      "available_at_launch": [
        "Consumer app",
        "Open weights"
      ],
      "summary": "An updated 350M-parameter model with pretraining extended from 10T to 28T tokens and large-scale reinforcement learning, aimed at tool use, data extraction, and structured outputs on CPUs, NPUs, and GPUs.",
      "highlights": [
        "Released as LFM2.5-350M-Base and post-trained LFM2.5-350M",
        "Available on Hugging Face, LEAP, and Liquid Playground",
        "New ecosystem partners at launch include Zetic, RunAnywhere, and Mirai"
      ],
      "source_title": "LFM2.5-350M: No Size Left Behind",
      "source_url": "https://www.liquid.ai/blog/lfm2-5-350m-no-size-left-behind",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#lfm2-5-350m"
    },
    {
      "slug": "veo-3-1-lite",
      "date": "2026-03-31",
      "model": "Veo 3.1 Lite",
      "lab": "Google",
      "lab_slug": "google",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms"
      ],
      "summary": "Veo 3.1 Lite is the most cost-effective model in Google's Veo 3.1 video generation lineup, aimed at high-volume developer use. It supports text-to-video and image-to-video at 720p or 1080p in landscape or portrait.",
      "highlights": [
        "Clip durations of 4, 6 or 8 seconds",
        "16:9 and 9:16 aspect ratios",
        "Paid preview in the Gemini API and Google AI Studio (veo-3.1-lite-generate-preview)",
        "Public preview on Vertex AI from April 2, 2026"
      ],
      "source_title": "Build with Veo 3.1 Lite, our most cost-effective video generation model",
      "source_url": "https://blog.google/innovation-and-ai/technology/ai/veo-3-1-lite/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#veo-3-1-lite"
    },
    {
      "slug": "longcat-audiodit",
      "date": "2026-03-30",
      "model": "LongCat-AudioDiT",
      "lab": "Meituan",
      "lab_slug": "meituan",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "LongCat-AudioDiT is a non-autoregressive diffusion text-to-speech model that operates directly in a waveform latent space, aimed at high-fidelity zero-shot voice cloning.",
      "highlights": [
        "Released in 1B and 3.5B sizes",
        "Diffusion TTS in the waveform latent space instead of mel-spectrograms",
        "Code and weights released under MIT"
      ],
      "source_title": "GitHub - meituan-longcat/LongCat-AudioDiT",
      "source_url": "https://github.com/meituan-longcat/LongCat-AudioDiT",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#longcat-audiodit"
    },
    {
      "slug": "qwen3-5-omni",
      "date": "2026-03-30",
      "model": "Qwen3.5-Omni",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": 256000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Qwen3.5-Omni is Qwen's omni-modal model series in Plus, Flash, and Light sizes, using a Thinker-Talker design for text and speech output. It adds semantic interruption, voice cloning, and coding from audio-visual instructions.",
      "highlights": [
        "Supports 256K long-context input",
        "Available through the Offline API and the Realtime API",
        "Talker generates contextual speech from the Thinker's outputs"
      ],
      "source_title": "Qwen3.5-Omni: Scaling Up, Toward Native Omni-Modal AGI",
      "source_url": "https://qwen.ai/blog?id=qwen3.5-omni",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#qwen3-5-omni"
    },
    {
      "slug": "granite-4-0-3b-vision",
      "date": "2026-03-27",
      "model": "Granite 4.0 3B Vision",
      "lab": "IBM",
      "lab_slug": "ibm",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Vision-language model from IBM designed for enterprise-grade document data extraction, focused on specialized, complex extraction tasks that ultracompact models often struggle with.",
      "highlights": [
        "Focused on document data extraction",
        "Predecessor to Granite Vision 4.1",
        "Released under Apache 2.0"
      ],
      "source_title": "ibm-granite/granite-4.0-3b-vision · Hugging Face",
      "source_url": "https://huggingface.co/ibm-granite/granite-4.0-3b-vision",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#granite-4-0-3b-vision"
    },
    {
      "slug": "meta-sam-3-1",
      "date": "2026-03-27",
      "model": "SAM 3.1",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "image",
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "SAM License",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "SAM 3.1 updates Segment Anything Model 3 with Object Multiplex, a shared-memory approach to joint multi-object tracking that makes real-time video detection and tracking significantly faster.",
      "highlights": [
        "Drop-in replacement checkpoint for SAM 3, published on Hugging Face",
        "Faster multi-object video tracking without sacrificing accuracy",
        "Offered as sam-3.1 in the Meta Model API"
      ],
      "source_title": "SAM 3.1: Faster and More Accessible Real-Time Video Detection and Tracking With Multiplexing and Global Reasoning",
      "source_url": "https://ai.meta.com/blog/segment-anything-model-3/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#meta-sam-3-1"
    },
    {
      "slug": "cohere-transcribe",
      "date": "2026-03-26",
      "model": "Cohere Transcribe",
      "lab": "Cohere",
      "lab_slug": "cohere",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "Cohere's first transcription model, an automatic speech recognition model that converts audio to text in 14 languages through the Audio Transcriptions API.",
      "highlights": [
        "Model name cohere-transcribe-03-2026",
        "Languages include English, German, French, Japanese, Korean, Chinese and Arabic",
        "Released under Apache 2.0"
      ],
      "source_title": "Announcing the Cohere Transcribe model",
      "source_url": "https://docs.cohere.com/v2/changelog/cohere-transcribe-03-2026",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#cohere-transcribe"
    },
    {
      "slug": "gemini-3-1-flash-live",
      "date": "2026-03-26",
      "model": "Gemini 3.1 Flash Live",
      "lab": "Google",
      "lab_slug": "google",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": 131072,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Gemini 3.1 Flash Live is Google's audio-to-audio model for real-time dialogue and voice-first applications, offered in preview in the Live API and powering Gemini Live and Search Live.",
      "highlights": [
        "Preview model gemini-3.1-flash-live-preview in the Gemini Live API via Google AI Studio",
        "Its multilingual support enabled Search Live's expansion to more than 200 countries and territories",
        "Input token limit of 131,072 per the Gemini API model page"
      ],
      "source_title": "Gemini 3.1 Flash Live: Google's latest AI audio model",
      "source_url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-flash-live/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#gemini-3-1-flash-live"
    },
    {
      "slug": "suno-v5-5",
      "date": "2026-03-26",
      "model": "Suno v5.5",
      "lab": "Suno",
      "lab_slug": "suno",
      "type": "music",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Suno's most expressive model at its release, launched alongside Voices, Custom Models and My Taste features for capturing a creator's voice and personal sound.",
      "highlights": [
        "Voices available to Pro and Premier subscribers",
        "Custom Models for a personalized sound",
        "My Taste personalization"
      ],
      "source_title": "Suno v5.5: More Expressive. More You.",
      "source_url": "https://suno.com/blog/v5-5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#suno-v5-5"
    },
    {
      "slug": "longcat-next",
      "date": "2026-03-25",
      "model": "LongCat-Next",
      "lab": "Meituan",
      "lab_slug": "meituan",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "audio"
      ],
      "output_modalities": [
        "text",
        "image",
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "LongCat-Next is a native multimodal model that processes text, vision, and audio under a single autoregressive objective, handling both understanding and generation in one model with discrete tokenizers.",
      "highlights": [
        "Described by Meituan as an A3B-size model (about 3B activated parameters)",
        "Discrete Native Autoregression (DiNA) paradigm for unified modalities",
        "Model weights and tokenizers released under MIT"
      ],
      "source_title": "GitHub - meituan-longcat/LongCat-Next",
      "source_url": "https://github.com/meituan-longcat/LongCat-Next",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#longcat-next"
    },
    {
      "slug": "lyria-3-pro",
      "date": "2026-03-25",
      "model": "Lyria 3 Pro",
      "lab": "Google",
      "lab_slug": "google",
      "type": "music",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API",
        "Cloud platforms"
      ],
      "summary": "Lyria 3 Pro is Google DeepMind's higher-tier Lyria 3 music model for full-length songs. It generates tracks up to three minutes long and understands song structure, so users can prompt for intros, verses, choruses and bridges.",
      "highlights": [
        "Tracks up to 3 minutes, versus 30 seconds for Lyria 3 in the Gemini app",
        "Public preview on Vertex AI; available in the Gemini API and AI Studio as lyria-3-pro-preview",
        "Longer generations in the Gemini app for paid subscribers, plus Google Vids and ProducerAI",
        "All outputs carry a SynthID watermark"
      ],
      "source_title": "Lyria 3 Pro: Create longer tracks in more Google products",
      "source_url": "https://blog.google/innovation-and-ai/technology/ai/lyria-3-pro/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#lyria-3-pro"
    },
    {
      "slug": "molmoweb",
      "date": "2026-03-24",
      "model": "MolmoWeb",
      "lab": "Ai2",
      "lab_slug": "ai2",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "4B and 8B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Open visual web agent built on Molmo 2 that navigates websites, fills out forms and retrieves information in a browser, designed for self-hosting.",
      "highlights": [
        "Released with weights, training data and evaluation tools",
        "Training data includes 36K human task trajectories",
        "Released under Apache 2.0"
      ],
      "source_title": "MolmoWeb: An open agent for automating web tasks | Ai2",
      "source_url": "https://allenai.org/blog/molmoweb",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#molmoweb"
    },
    {
      "slug": "voxtral-tts",
      "date": "2026-03-23",
      "model": "Voxtral TTS",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": "4B",
      "model_license": "CC-BY-NC-4.0",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Mistral's first text-to-speech model, a 4B model for expressive multilingual speech in nine languages with zero-shot voice adaptation from a short voice prompt.",
      "highlights": [
        "Streaming with about 90 ms time to first audio",
        "Weights with reference voices released under CC BY-NC 4.0",
        "API model ID voxtral-mini-tts-2603"
      ],
      "source_title": "Speaking of Voxtral | Mistral AI",
      "source_url": "https://mistral.ai/news/voxtral-tts",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#voxtral-tts"
    },
    {
      "slug": "longcat-flash-prover",
      "date": "2026-03-20",
      "model": "LongCat-Flash-Prover",
      "lab": "Meituan",
      "lab_slug": "meituan",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "MoE: 560B total",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "LongCat-Flash-Prover is a 560B-parameter mixture-of-experts model for formal reasoning in Lean 4, using agentic tool-integrated reasoning for auto-formalization, proof sketching, and theorem proving.",
      "highlights": [
        "Targets auto-formalization, lemma-style sketching, and whole-proof generation",
        "Trained with the Hierarchical Importance Sampling Policy Optimization (HisPO) algorithm",
        "Weights released under MIT"
      ],
      "source_title": "GitHub - meituan-longcat/LongCat-Flash-Prover",
      "source_url": "https://github.com/meituan-longcat/LongCat-Flash-Prover",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#longcat-flash-prover"
    },
    {
      "slug": "mai-image-2",
      "date": "2026-03-19",
      "model": "MAI-Image-2",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "MAI-Image-2 is Microsoft AI's second in-house text-to-image model, focused on photorealism, reliable in-image text and detailed scene generation. It launched in MAI Playground and began rolling out to Copilot and Bing Image Creator.",
      "highlights": [
        "Available in MAI Playground at launch",
        "API access first for select customers, then in Microsoft Foundry from April 2, 2026",
        "Rolling out to Copilot and Bing Image Creator"
      ],
      "source_title": "Introducing MAI-Image-2: for limitless creativity",
      "source_url": "https://microsoft.ai/news/introducing-mai-image-2/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#mai-image-2"
    },
    {
      "slug": "mimo-v2-omni",
      "date": "2026-03-18",
      "model": "MiMo-V2-Omni",
      "lab": "Xiaomi",
      "lab_slug": "xiaomi",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video",
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "MiMo-V2-Omni is Xiaomi's omnimodal agentic foundation model, fusing dedicated image, video, and audio encoders into a shared backbone so it can perceive several modalities at once and act on them.",
      "highlights": [
        "Supports up to 256K tokens of context",
        "Trained to connect perception with next actions for agent use",
        "Available through the MiMo API Platform"
      ],
      "source_title": "MiMo-V2-Omni | Xiaomi",
      "source_url": "https://mimo.xiaomi.com/mimo-v2-omni/index.html",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#mimo-v2-omni"
    },
    {
      "slug": "mimo-v2-pro",
      "date": "2026-03-18",
      "model": "MiMo-V2-Pro",
      "lab": "Xiaomi",
      "lab_slug": "xiaomi",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": "MoE: over 1T total, 42B active",
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "MiMo-V2-Pro is Xiaomi's flagship foundation model for agentic workloads, scaled to more than 1T total parameters and designed to orchestrate complex workflows and production engineering tasks.",
      "highlights": [
        "Hybrid attention with a 7:1 ratio of sliding-window to global attention",
        "Supports up to 1M tokens of context",
        "Lightweight multi-token prediction layer for faster generation",
        "Released alongside MiMo-V2-Omni and MiMo-V2-TTS"
      ],
      "source_title": "MiMo-V2-Pro | Xiaomi",
      "source_url": "https://mimo.xiaomi.com/mimo-v2-pro/index.html",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#mimo-v2-pro"
    },
    {
      "slug": "mimo-v2-tts",
      "date": "2026-03-18",
      "model": "MiMo-V2-TTS",
      "lab": "Xiaomi",
      "lab_slug": "xiaomi",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "MiMo-V2-TTS is Xiaomi's large-scale speech synthesis model, built on an in-house audio tokenizer and multi-codebook speech-text modeling and pretrained on more than 100 million hours of speech data.",
      "highlights": [
        "Supports style control, singing, and voice cloning",
        "Refined with multi-dimensional reinforcement learning",
        "Launch demo covers Northeastern Mandarin, Sichuanese, and Cantonese speech"
      ],
      "source_title": "MiMo-V2-TTS | Xiaomi",
      "source_url": "https://mimo.xiaomi.com/mimo-v2-tts/index.html",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#mimo-v2-tts"
    },
    {
      "slug": "minimax-m2-7",
      "date": "2026-03-18",
      "model": "MiniMax M2.7",
      "lab": "MiniMax",
      "lab_slug": "minimax",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Non-Commercial License",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "MiniMax M2.7 is a language model for software engineering and professional office work that MiniMax describes as its first model deeply involved in its own development, building the agent scaffolding, skills, and memory used in its RL experiments.",
      "highlights": [
        "Released on the API as MiniMax-M2.7 and M2.7-highspeed",
        "Weights open-sourced on April 12, 2026",
        "Weights license permits non-commercial use; commercial use requires written authorization",
        "Uses Agent Teams, complex Skills, and dynamic tool search"
      ],
      "source_title": "MiniMax M2.7: Early Echoes of Self-Evolution",
      "source_url": "https://www.minimax.io/news/minimax-m27-en",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#minimax-m2-7"
    },
    {
      "slug": "qianfan-ocr",
      "date": "2026-03-18",
      "model": "Qianfan-OCR",
      "lab": "Baidu",
      "lab_slug": "baidu",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "image",
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "4B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Qianfan-OCR is a 4B end-to-end document intelligence model from Baidu's Qianfan team that unifies document parsing, layout analysis, table and formula recognition, chart understanding, and key information extraction in one model.",
      "highlights": [
        "Optional Layout-as-Thought phase reasons about bounding boxes, element types, and reading order before output",
        "Language model based on Qwen3-4B with 32K context, extendable to 131K",
        "Supports 192 languages",
        "Trained on 1,024 Kunlun P800 chips"
      ],
      "source_title": "GitHub - baidubce/Qianfan-VL: Qianfan-VL: Domain-Enhanced Universal Vision-Language Models",
      "source_url": "https://github.com/baidubce/Qianfan-VL",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#qianfan-ocr"
    },
    {
      "slug": "gpt-5-4-mini",
      "date": "2026-03-17",
      "model": "GPT-5.4 mini and GPT-5.4 nano",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "small",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 400000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "GPT-5.4 mini and nano are smaller GPT-5.4 models for high-volume workloads; mini targets coding, subagents, and computer use at more than twice GPT-5 mini's speed, while nano handles classification, extraction, and ranking.",
      "highlights": [
        "API model IDs gpt-5.4-mini and gpt-5.4-nano; both list a 400,000-token context window",
        "Mini is in the API, Codex, and ChatGPT (Free and Go via Thinking; rate-limit fallback for others)",
        "Nano is API-only and supports compaction but not tool search or computer use",
        "Mini supports tool search, built-in computer use, and compaction"
      ],
      "source_title": "Introducing GPT-5.4 mini and nano | OpenAI",
      "source_url": "https://openai.com/index/introducing-gpt-5-4-mini-and-nano/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#gpt-5-4-mini"
    },
    {
      "slug": "midjourney-v8-alpha",
      "date": "2026-03-17",
      "model": "Midjourney V8 (Alpha)",
      "lab": "Midjourney",
      "lab_slug": "midjourney",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Early V8 image model opened to the community on alpha.midjourney.com, with better adherence to detailed directions, improved text rendering, more coherent images and roughly five times faster generation.",
      "highlights": [
        "New --hd mode renders natively at 2K",
        "Backward compatible with V7 personalization profiles, moodboards and srefs",
        "Scheduled for deprecation two weeks after V8.1 became the default"
      ],
      "source_title": "V8 Alpha",
      "source_url": "https://updates.midjourney.com/v8-alpha/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#midjourney-v8-alpha"
    },
    {
      "slug": "leanstral",
      "date": "2026-03-16",
      "model": "Leanstral",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "MoE: 119B total, 6.5B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Mistral's first open-source code agent for the Lean 4 proof assistant, a highly sparse model built for formal proof engineering in realistic repositories rather than isolated math problems.",
      "highlights": [
        "Available in an agent mode within Mistral Vibe and through an API endpoint",
        "Announced with a new FLTEval evaluation suite for proof engineering",
        "API model ID labs-leanstral-2603"
      ],
      "source_title": "Leanstral: Open-Source foundation for trustworthy vibe-coding | Mistral AI",
      "source_url": "https://mistral.ai/news/leanstral",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#leanstral"
    },
    {
      "slug": "mistral-small-4",
      "date": "2026-03-16",
      "model": "Mistral Small 4",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "MoE: 119B total, 6B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Hybrid model that unifies the reasoning of Magistral, the vision of Pixtral and the agentic coding of Devstral in one open-weight model with configurable reasoning effort.",
      "highlights": [
        "128 experts with 4 active per token",
        "Released as Mistral joined the NVIDIA Nemotron Coalition as a founding member",
        "API model ID mistral-small-2603"
      ],
      "source_title": "Introducing Mistral Small 4 | Mistral AI",
      "source_url": "https://mistral.ai/news/mistral-small-4",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#mistral-small-4"
    },
    {
      "slug": "nemotron-3-nano-4b",
      "date": "2026-03-16",
      "model": "Nemotron 3 Nano 4B",
      "lab": "NVIDIA",
      "lab_slug": "nvidia",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 262000,
      "parameters": "3.97B",
      "model_license": "NVIDIA Nemotron Open Model License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Compact member of the Nemotron 3 family with about 4B parameters and a long context window, for local and edge deployments.",
      "highlights": [
        "Context length up to 262K tokens",
        "Released under the NVIDIA Nemotron Open Model License",
        "Smaller sibling of Nemotron 3 Nano 30B"
      ],
      "source_title": "nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16 · Hugging Face",
      "source_url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#nemotron-3-nano-4b"
    },
    {
      "slug": "glm-5-turbo",
      "date": "2026-03-15",
      "model": "GLM-5-Turbo",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 200000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "GLM-5-Turbo is a GLM-5-based model that Z.ai optimized for OpenClaw-style agent workloads. It strengthens calls to external tools and skills and is more stable on long, multi-step tasks.",
      "highlights": [
        "Improved decomposition of complex instructions and multi-agent coordination",
        "Up to 128K output tokens",
        "Available through the BigModel and Z.ai APIs"
      ],
      "source_title": "Model and product release records (BigModel docs)",
      "source_url": "https://docs.bigmodel.cn/cn/update/new-releases",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#glm-5-turbo"
    },
    {
      "slug": "mistral-moderation-2",
      "date": "2026-03-12",
      "model": "Mistral Moderation 2",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Second-generation Mistral moderation model with a 128K-token context window and jailbreak detection, built for complex multilingual content including long multi-turn conversations.",
      "highlights": [
        "API model ID mistral-moderation-2603",
        "Released with custom guardrails for Agents and Conversations",
        "Mistral's model card lists March 1, 2026"
      ],
      "source_title": "Changelog | Mistral Docs",
      "source_url": "https://docs.mistral.ai/resources/changelogs",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#mistral-moderation-2"
    },
    {
      "slug": "longcat-flash-omni-2603",
      "date": "2026-03-11",
      "model": "LongCat-Flash-Omni-2603",
      "lab": "Meituan",
      "lab_slug": "meituan",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "LongCat-Flash-Omni-2603 is an upgraded version of LongCat-Flash-Omni, an end-to-end omni-modal interaction model with more natural spoken dialogue and stronger perception across vision, speech, and text.",
      "highlights": [
        "Available in LongCat Chat and on the API as model=LongCat-Flash-Omni-2603",
        "Adds native voice function calling from audio commands",
        "Retired from the LongCat API on May 29, 2026"
      ],
      "source_title": "Change Log | LongCat API Docs",
      "source_url": "https://longcat.chat/platform/docs/change-log",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#longcat-flash-omni-2603"
    },
    {
      "slug": "nemotron-3-super",
      "date": "2026-03-11",
      "model": "Nemotron 3 Super",
      "lab": "NVIDIA",
      "lab_slug": "nvidia",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": "MoE: 120B total, 12B active",
      "model_license": "NVIDIA Nemotron Open Model License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Mid-size Nemotron 3 model built on a hybrid latent mixture-of-experts architecture with Mamba-2, MoE and attention layers, aimed at agentic workflows and long-context reasoning.",
      "highlights": [
        "Adds multi-token prediction layers",
        "Supports up to 1M tokens of context",
        "Trained between December 2025 and March 2026"
      ],
      "source_title": "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16 · Hugging Face",
      "source_url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#nemotron-3-super"
    },
    {
      "slug": "reka-edge-2603",
      "date": "2026-03-11",
      "model": "Reka Edge (reka-edge-2603)",
      "lab": "Reka",
      "lab_slug": "reka",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 16384,
      "parameters": "7B (657M vision encoder, 6.4B transformer)",
      "model_license": "Business Source License 1.1 (Reka Edge 2603 License)",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "A new version of Reka's 7B vision-language model for physical AI and edge devices, pairing a ConvNeXt V2 vision encoder with a transformer trained from scratch and using 64 tokens per image tile.",
      "highlights": [
        "API model ID reka-edge-2603 with a 16,384-token context per Reka docs",
        "Targets image understanding, video analysis, object detection and agentic tool use",
        "Open weights on Hugging Face; the license permits commercial use below an annual revenue threshold",
        "Runs on NVIDIA Jetson, Apple Silicon Macs, PCs and Snapdragon devices; vLLM plugin available"
      ],
      "source_title": "Reka Edge: Frontier-Level Edge Intelligence for Physical AI",
      "source_url": "https://reka.ai/news/reka-edge-frontier-level-edge-intelligence-for-physical-ai",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#reka-edge-2603"
    },
    {
      "slug": "gemini-embedding-2",
      "date": "2026-03-10",
      "model": "Gemini Embedding 2",
      "lab": "Google",
      "lab_slug": "google",
      "type": "embedding",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video",
        "audio"
      ],
      "output_modalities": [
        "embedding"
      ],
      "context_window": 8192,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms"
      ],
      "summary": "Gemini Embedding 2 is Google's first natively multimodal embedding model, mapping text, images, video, audio and documents into a single embedding space for multimodal retrieval and classification.",
      "highlights": [
        "Released in public preview in the Gemini API and Vertex AI; generally available on Apr 22, 2026",
        "Accepts interleaved multimodal input and captures semantic intent across over 100 languages",
        "Default 3072-dimension output with Matryoshka Representation Learning for smaller sizes",
        "Input token limit of 8,192 per the Gemini API model page"
      ],
      "source_title": "Gemini Embedding 2: Our first natively multimodal embedding model",
      "source_url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-embedding-2/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#gemini-embedding-2"
    },
    {
      "slug": "grok-4-20",
      "date": "2026-03-10",
      "model": "Grok 4.20 and Grok 4.20 Multi-agent",
      "lab": "xAI",
      "lab_slug": "xai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Grok 4.20 is a model family on the xAI API with reasoning and non-reasoning variants, plus a Multi-agent version in which several agents collaborate on research tasks and a leader agent synthesizes the answer.",
      "highlights": [
        "API IDs: grok-4.20-0309-reasoning, grok-4.20-0309-non-reasoning, grok-4.20-multi-agent-0309",
        "1M-token context window with text and image input",
        "Multi-agent reasoning effort sets the number of agents (4 or 16)"
      ],
      "source_title": "Release Notes | SpaceXSI Docs",
      "source_url": "https://docs.x.ai/developers/release-notes",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#grok-4-20"
    },
    {
      "slug": "runway-characters",
      "date": "2026-03-09",
      "model": "Runway Characters (GWM-1 Avatars)",
      "lab": "Runway",
      "lab_slug": "runway",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "image",
        "audio",
        "text"
      ],
      "output_modalities": [
        "video",
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Enterprise"
      ],
      "summary": "Real-time video agent API for fully custom conversational characters built on the GWM-1 world model, generating expressive avatars from a single image with control over voice and personality.",
      "highlights": [
        "Built on the GWM-1 general world model announced December 2025",
        "Zero fine-tuning; any visual style from one image",
        "Available via the Runway API with a web demo"
      ],
      "source_title": "Introducing Runway Characters",
      "source_url": "https://runway.com/news/introducing-runway-characters",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#runway-characters"
    },
    {
      "slug": "granite-4-0-1b-speech",
      "date": "2026-03-06",
      "model": "Granite 4.0 1B Speech",
      "lab": "IBM",
      "lab_slug": "ibm",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Compact speech-language model for multilingual automatic speech recognition and bidirectional speech translation, with half the parameters of Granite Speech 3.3 2B.",
      "highlights": [
        "Built for resource-constrained devices",
        "Supports Japanese and keyword-biased recognition",
        "Released under Apache 2.0"
      ],
      "source_title": "ibm-granite/granite-4.0-1b-speech · Hugging Face",
      "source_url": "https://huggingface.co/ibm-granite/granite-4.0-1b-speech",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#granite-4-0-1b-speech"
    },
    {
      "slug": "gpt-5-4",
      "date": "2026-03-05",
      "model": "GPT-5.4 (Thinking and Pro)",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1050000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API",
        "Enterprise"
      ],
      "summary": "GPT-5.4 is a frontier model for professional work that folds in GPT-5.3-Codex coding capabilities, adds native computer use and tool search in the API, and ships in ChatGPT as GPT-5.4 Thinking alongside GPT-5.4 Pro.",
      "highlights": [
        "API model IDs gpt-5.4 and gpt-5.4-pro (Pro in the Responses API only)",
        "1,050,000-token context window per the API model page, with native compaction support",
        "GPT-5.4 Thinking for Plus, Team, and Pro, replacing GPT-5.2 Thinking; GPT-5.4 Pro for Pro and Enterprise",
        "Also rolled out in Codex, with experimental 1M-token context support"
      ],
      "source_title": "Introducing GPT-5.4 | OpenAI",
      "source_url": "https://openai.com/index/introducing-gpt-5-4/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#gpt-5-4"
    },
    {
      "slug": "ltx-2-3",
      "date": "2026-03-05",
      "model": "LTX-2.3",
      "lab": "Lightricks",
      "lab_slug": "lightricks",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video",
        "audio"
      ],
      "output_modalities": [
        "video",
        "audio"
      ],
      "context_window": null,
      "parameters": "22B",
      "model_license": "LTX-2 Community License",
      "available_at_launch": [
        "Open weights",
        "API",
        "Consumer app"
      ],
      "summary": "Upgraded LTX-2 model with a 22B-parameter architecture, sharper details from a new VAE, cleaner audio, stronger image-to-video and better prompt understanding, offered as Pro and Fast API variants.",
      "highlights": [
        "22B parameters, up from 19B",
        "Up to 4K at 50 fps and up to 20 seconds per clip",
        "Native 9:16 portrait video"
      ],
      "source_title": "Release Notes | LTX",
      "source_url": "https://ltx.io/release-notes",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#ltx-2-3"
    },
    {
      "slug": "olmo-hybrid",
      "date": "2026-03-05",
      "model": "Olmo Hybrid",
      "lab": "Ai2",
      "lab_slug": "ai2",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "7B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Fully open 7B model family that mixes transformer attention with linear recurrent layers, released with a controlled comparison against Olmo 3 7B.",
      "highlights": [
        "Combines transformers and linear RNNs",
        "Developed through 1B and 7B scale experiments",
        "Released under Apache 2.0"
      ],
      "source_title": "Introducing Olmo Hybrid: Combining transformers and linear RNNs for superior scaling | Ai2",
      "source_url": "https://allenai.org/blog/olmohybrid",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#olmo-hybrid"
    },
    {
      "slug": "minimax-music-2-5-plus",
      "date": "2026-03-04",
      "model": "MiniMax Music 2.5+",
      "lab": "MiniMax",
      "lab_slug": "minimax",
      "type": "music",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Music 2.5+ extends MiniMax Music 2.5 with instrumental music generation, producing tracks without vocals in styles from classical orchestration and modern electronic to ambient sound and natural soundscapes.",
      "highlights": [
        "Listed in the API release notes as Music2.5+",
        "Targets uses such as meditation, sleep aids, advertising, and game scoring",
        "Style switching requires no additional tuning"
      ],
      "source_title": "Music 2.5+: Unlock instrumental music, break through style boundaries",
      "source_url": "https://www.minimax.io/news/music-25-unlock-instrumental-music",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#minimax-music-2-5-plus"
    },
    {
      "slug": "phi-4-reasoning-vision-15b",
      "date": "2026-03-04",
      "model": "Phi-4-reasoning-vision-15B",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 16384,
      "parameters": "15B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Phi-4-reasoning-vision-15B is Microsoft's open-weight multimodal reasoning model, pairing a SigLIP-2 vision encoder with the Phi-4-Reasoning backbone. It reasons step by step for math and science and answers perception tasks directly.",
      "highlights": [
        "15B parameters with a 16,384-token context length",
        "Built on the Phi-4-Reasoning backbone with a SigLIP-2 vision encoder",
        "Available on Microsoft Foundry, Hugging Face and GitHub"
      ],
      "source_title": "Phi-4-reasoning-vision and the lessons of training a multimodal reasoning model",
      "source_url": "https://www.microsoft.com/en-us/research/blog/phi-4-reasoning-vision-and-the-lessons-of-training-a-multimodal-reasoning-model/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#phi-4-reasoning-vision-15b"
    },
    {
      "slug": "gemini-3-1-flash-lite",
      "date": "2026-03-03",
      "model": "Gemini 3.1 Flash-Lite",
      "lab": "Google",
      "lab_slug": "google",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1048576,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms"
      ],
      "summary": "Gemini 3.1 Flash-Lite is the first Flash-Lite model in the Gemini 3 series, Google's fastest and most cost-efficient Gemini 3 model for high-volume workloads such as translation and content moderation.",
      "highlights": [
        "Preview in the Gemini API via Google AI Studio and in Vertex AI at launch",
        "Stable gemini-3.1-flash-lite became generally available on May 7, 2026",
        "Input token limit of 1,048,576 per the Gemini API model page"
      ],
      "source_title": "Gemini 3.1 Flash Lite: Our most cost-effective AI model yet",
      "source_url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-flash-lite/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#gemini-3-1-flash-lite"
    },
    {
      "slug": "gpt-5-3-instant",
      "date": "2026-03-03",
      "model": "GPT-5.3 Instant",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "GPT-5.3 Instant is an update to ChatGPT's most-used model, aimed at fewer unnecessary refusals and caveats, better-synthesized answers from web search, a more direct conversational style, and stronger writing.",
      "highlights": [
        "Available to all ChatGPT users and in the API as gpt-5.3-chat-latest",
        "Replaced GPT-5.2 Instant, which was retired from ChatGPT on June 3, 2026",
        "128,000-token context window per the gpt-5.3-chat-latest API model page",
        "A March 16, 2026 update improved follow-up tone and reduced teaser-style phrasing"
      ],
      "source_title": "GPT-5.3 Instant: Smoother, more useful everyday conversations | OpenAI",
      "source_url": "https://openai.com/index/gpt-5-3-instant/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#gpt-5-3-instant"
    },
    {
      "slug": "qwen3-5-small",
      "date": "2026-03-02",
      "model": "Qwen3.5 small models",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 262144,
      "parameters": "0.8B, 2B, 4B, 9B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "The Qwen3.5 small models are four compact open-weights models, Qwen3.5-9B, 4B, 2B, and 0.8B, completing the Qwen3.5 lineup. Like the rest of the series, they are causal language models with a vision encoder.",
      "highlights": [
        "Native 262,144-token context, extensible to 1,010,000 tokens",
        "Base checkpoints released alongside the instruct models",
        "Weights on Hugging Face and ModelScope under the Apache 2.0 license"
      ],
      "source_title": "Qwen3.8 (GitHub), news entry for Qwen3.5 small models",
      "source_url": "https://github.com/QwenLM/Qwen3.8",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/03#qwen3-5-small"
    },
    {
      "slug": "nano-banana-2",
      "date": "2026-02-26",
      "model": "Nano Banana 2 (Gemini 3.1 Flash Image)",
      "lab": "Google",
      "lab_slug": "google",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text",
        "image"
      ],
      "context_window": 131072,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms",
        "Consumer app"
      ],
      "summary": "Nano Banana 2 (Gemini 3.1 Flash Image) is a high-efficiency image generation and editing model that brings Nano Banana Pro's world knowledge and reasoning to Flash-level speed.",
      "highlights": [
        "Supports aspect ratios and resolutions from 512px to 4K",
        "Replaced Nano Banana Pro as the default image model across the Gemini app's Fast, Thinking and Pro models",
        "Stable gemini-3.1-flash-image became generally available on May 28, 2026, adding video-to-image generation"
      ],
      "source_title": "Nano Banana 2: Google's latest AI image generation model",
      "source_url": "https://blog.google/innovation-and-ai/technology/ai/nano-banana-2/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/02#nano-banana-2"
    },
    {
      "slug": "pplx-embed-v1",
      "date": "2026-02-26",
      "model": "pplx-embed-v1 and pplx-embed-context-v1",
      "lab": "Perplexity",
      "lab_slug": "perplexity",
      "type": "embedding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "embedding"
      ],
      "context_window": null,
      "parameters": "0.6B and 4B",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "Two text embedding model families, pplx-embed-v1 for standard dense retrieval and pplx-embed-context-v1 for document-aware chunk embeddings, each in 0.6B and 4B sizes and built by diffusion continued pretraining of Qwen3.",
      "highlights": [
        "32K-token context, native INT8 and binary embeddings, Matryoshka dimensions, no instruction prefixes",
        "API model IDs pplx-embed-v1-0.6b, pplx-embed-v1-4b, pplx-embed-context-v1-0.6b and pplx-embed-context-v1-4b",
        "Perplexity's Embeddings API became available the same month (February 2026)",
        "Weights on Hugging Face under the MIT license"
      ],
      "source_title": "pplx-embed: State-of-the-Art Embedding Models for Web-Scale Retrieval",
      "source_url": "https://www.perplexity.ai/hub/blog/pplx-embed-state-of-the-art-embedding-models-for-web-scale-retrieval",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/02#pplx-embed-v1"
    },
    {
      "slug": "lfm2-24b-a2b",
      "date": "2026-02-24",
      "model": "LFM2-24B-A2B",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32768,
      "parameters": "MoE: 24B total, 2.3B active",
      "model_license": "LFM Open License v1.0",
      "available_at_launch": [
        "Consumer app",
        "Open weights"
      ],
      "summary": "An early checkpoint of the largest LFM2 model, a sparse mixture-of-experts instruct model with 24B total and 2.3B active parameters, designed to fit in 32GB of RAM for laptops, desktops, and cloud deployment.",
      "highlights": [
        "40 layers with 64 experts per MoE block and top-4 routing",
        "Instruct model without reasoning traces; trained on 17T tokens at release",
        "Day-zero support in llama.cpp, vLLM, and SGLang",
        "Weights on Hugging Face; testable on Liquid Playground"
      ],
      "source_title": "LFM2-24B-A2B: Scaling Up the LFM2 Architecture",
      "source_url": "https://www.liquid.ai/blog/lfm2-24b-a2b",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/02#lfm2-24b-a2b"
    },
    {
      "slug": "qwen3-5-medium",
      "date": "2026-02-24",
      "model": "Qwen3.5 medium models",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 262144,
      "parameters": "MoE: 122B total, 10B active; MoE: 35B total, 3B active; 27B dense",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "The Qwen3.5 medium models are Qwen3.5-122B-A10B, Qwen3.5-35B-A3B, and the dense Qwen3.5-27B, released after the flagship Qwen3.5-397B-A17B. They share the Qwen3.5 hybrid architecture with native vision.",
      "highlights": [
        "Native 262,144-token context, extensible to 1,010,000 tokens",
        "Qwen3.5-35B-A3B also released as a base model",
        "Weights under the Apache 2.0 license"
      ],
      "source_title": "Qwen3.8 (GitHub), news entry for Qwen3.5 medium models",
      "source_url": "https://github.com/QwenLM/Qwen3.8",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/02#qwen3-5-medium"
    },
    {
      "slug": "gpt-realtime-1-5",
      "date": "2026-02-23",
      "model": "gpt-realtime-1.5 and gpt-audio-1.5",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "audio",
        "image"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "GPT-Realtime-1.5 is an updated speech-to-speech model for voice agents and customer support in the Realtime API, released with gpt-audio-1.5, an audio-in, audio-out model for the Chat Completions API.",
      "highlights": [
        "API model IDs gpt-realtime-1.5 (v1/realtime) and gpt-audio-1.5 (v1/chat/completions)",
        "gpt-realtime-1.5 lists a 32,000-token context window; gpt-audio-1.5 lists 128,000 tokens",
        "gpt-realtime-1.5 accepts image input in addition to text and audio"
      ],
      "source_title": "Changelog | OpenAI API",
      "source_url": "https://developers.openai.com/api/docs/changelog",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/02#gpt-realtime-1-5"
    },
    {
      "slug": "gemini-3-1-pro",
      "date": "2026-02-19",
      "model": "Gemini 3.1 Pro",
      "lab": "Google",
      "lab_slug": "google",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1048576,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms",
        "Consumer app"
      ],
      "summary": "Gemini 3.1 Pro is an upgraded Gemini 3 Pro that carries the core intelligence behind the February 2026 Deep Think update, released in preview across the Gemini API, Vertex AI, the Gemini app and NotebookLM.",
      "highlights": [
        "Preview model gemini-3.1-pro-preview, plus a gemini-3.1-pro-preview-customtools endpoint that prioritizes custom tools",
        "Rolled out with higher limits for Google AI Pro and Ultra subscribers in the Gemini app",
        "Input token limit of 1,048,576 per the Gemini API model page"
      ],
      "source_title": "Gemini 3.1 Pro: Announcing our latest Gemini AI model",
      "source_url": "https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-pro/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/02#gemini-3-1-pro"
    },
    {
      "slug": "lyria-3",
      "date": "2026-02-18",
      "model": "Lyria 3",
      "lab": "Google",
      "lab_slug": "google",
      "type": "music",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Lyria 3 is a Google DeepMind music generation model that launched in beta in the Gemini app. It creates 30-second tracks with automatically written lyrics and cover art from a text description, photo or video.",
      "highlights": [
        "Available to Gemini app users 18+ in eight languages, with higher limits for paid plans",
        "Also used in YouTube Dream Track for Shorts soundtracks",
        "Every Gemini-generated track carries a SynthID watermark",
        "Reached the Gemini API as lyria-3-clip-preview on March 25, 2026"
      ],
      "source_title": "A new way to express yourself: Gemini can now create music",
      "source_url": "https://blog.google/innovation-and-ai/products/gemini-app/lyria-3/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/02#lyria-3"
    },
    {
      "slug": "claude-sonnet-4-6",
      "date": "2026-02-17",
      "model": "Claude Sonnet 4.6",
      "lab": "Anthropic",
      "lab_slug": "anthropic",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Cloud platforms"
      ],
      "summary": "Claude Sonnet 4.6 is a full upgrade of Anthropic's Sonnet model across coding, computer use, long-context reasoning, agent planning, knowledge work, and design. It became the default model for Free and Pro users on claude.ai.",
      "highlights": [
        "API model ID claude-sonnet-4-6",
        "1M-token context window launched in beta; out of beta on March 13, 2026",
        "Supports both adaptive thinking and extended thinking",
        "Available on all Claude plans, the Claude API, and all major cloud platforms"
      ],
      "source_title": "Introducing Claude Sonnet 4.6",
      "source_url": "https://www.anthropic.com/news/claude-sonnet-4-6",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/02#claude-sonnet-4-6"
    },
    {
      "slug": "recraft-v4",
      "date": "2026-02-17",
      "model": "Recraft V4",
      "lab": "Recraft",
      "lab_slug": "recraft",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Ground-up rebuild of Recraft's image model tuned with designers for visual taste, prompt accuracy and output quality, offered as V4 and a higher-resolution V4 Pro.",
      "highlights": [
        "V4 Pro generates higher-resolution, print-ready images",
        "V4 Vector and V4 Pro Vector produce editable SVG files",
        "Available in Recraft and via API"
      ],
      "source_title": "Introducing Recraft V4",
      "source_url": "https://www.recraft.ai/blog/introducing-recraft-v4-design-taste-meets-image-generation",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/02#recraft-v4"
    },
    {
      "slug": "tiny-aya",
      "date": "2026-02-17",
      "model": "Tiny Aya",
      "lab": "Cohere",
      "lab_slug": "cohere",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "3.35B",
      "model_license": "CC-BY-NC-4.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Compact multilingual open-weight models from Cohere Labs that are small enough to run locally on consumer hardware and phones, with a 3.35B base covering 70+ languages.",
      "highlights": [
        "TinyAya-Global instruct model supports 67 languages",
        "Regional variants Earth, Fire and Water published on Hugging Face",
        "Released with a new multilingual fine-tuning dataset and benchmarks"
      ],
      "source_title": "Cohere Labs Launches Tiny Aya for Multilingual AI | Cohere",
      "source_url": "https://cohere.com/blog/cohere-labs-tiny-aya",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/02#tiny-aya"
    },
    {
      "slug": "qwen3-5",
      "date": "2026-02-16",
      "model": "Qwen3.5",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 262144,
      "parameters": "MoE: 397B total, 17B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "Qwen3.5 is Alibaba's model series that opened with the open-weights Qwen3.5-397B-A17B, a native vision-language model. It uses a hybrid architecture that combines linear attention through Gated Delta Networks with a sparse mixture-of-experts design.",
      "highlights": [
        "Native 262,144-token context, extensible to 1,010,000 tokens",
        "Language and dialect support expanded from 119 to 201",
        "Hosted Qwen3.5-Plus offers a 1M context window by default with built-in tools",
        "Weights under the Apache 2.0 license"
      ],
      "source_title": "Qwen3.5: Towards Native Multimodal Agents",
      "source_url": "https://qwen.ai/blog?id=qwen3.5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/02#qwen3-5"
    },
    {
      "slug": "seed-2-0",
      "date": "2026-02-14",
      "model": "Seed2.0",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Seed2.0 is ByteDance Seed's model series for complex real-world tasks, offered as Pro, Lite, and Mini general agent models plus a dedicated Code model, with stronger visual understanding, long-context handling, and multi-step instruction following.",
      "highlights": [
        "Three general agent sizes (Pro, Lite, Mini) plus a Code model",
        "Pro and Code models launched in the Doubao app and TRAE",
        "Full series available through the Volcano Engine API",
        "Upgraded document, chart, and long-video understanding over Seed1.8"
      ],
      "source_title": "Seed 2.0 Official Launch",
      "source_url": "https://seed.bytedance.com/en/blog/seed-2-0-official-launch",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/02#seed-2-0"
    },
    {
      "slug": "seedream-5-0-lite",
      "date": "2026-02-13",
      "model": "Seedream 5.0 Lite",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Seedream 5.0 Lite is an image creation model that brings deeper reasoning and world knowledge to generation and editing, and adds real-time web search for time-sensitive content, building on Seedream 4.0 and 4.5.",
      "highlights": [
        "Adds real-time search enhancement to image generation",
        "Improved knowledge reasoning, editing response, and consistency",
        "Launched on Jimeng AI and Volcano Ark's experience center, with Doubao in internal testing"
      ],
      "source_title": "Deeper Thinking, More Accurate Generation | Introducing Seedream 5.0 Lite",
      "source_url": "https://seed.bytedance.com/en/blog/deeper-thinking-more-accurate-generation-introducing-seedream-5-0-lite",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/02#seedream-5-0-lite"
    },
    {
      "slug": "glm-5",
      "date": "2026-02-12",
      "model": "GLM-5",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 200000,
      "parameters": "MoE: 744B total, 40B active",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "GLM-5 is Z.ai's open-weights model for complex systems engineering and long-running agent tasks. It scales to 744B total parameters from 355B in GLM-4.5 and integrates DeepSeek Sparse Attention to reduce deployment cost.",
      "highlights": [
        "Pre-training data increased from 23T to 28.5T tokens",
        "Trained with slime, Z.ai's asynchronous RL infrastructure",
        "Up to 128K output tokens",
        "Weights under the MIT license"
      ],
      "source_title": "GLM-5: From Vibe Coding to Agentic Engineering",
      "source_url": "https://z.ai/blog/glm-5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/02#glm-5"
    },
    {
      "slug": "gpt-5-3-codex-spark",
      "date": "2026-02-12",
      "model": "GPT-5.3-Codex-Spark",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "coding",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "GPT-5.3-Codex-Spark is a research preview of a smaller GPT-5.3-Codex designed for real-time coding in Codex, served on Cerebras hardware at more than 1,000 tokens per second.",
      "highlights": [
        "Research preview for ChatGPT Pro users in the Codex app, CLI, and VS Code extension",
        "Text-only at launch with a 128K context window",
        "Runs on the Cerebras Wafer Scale Engine 3, the first milestone of the OpenAI and Cerebras partnership",
        "API access limited to a small set of design partners at launch"
      ],
      "source_title": "Introducing GPT-5.3-Codex-Spark | OpenAI",
      "source_url": "https://openai.com/index/introducing-gpt-5-3-codex-spark/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/02#gpt-5-3-codex-spark"
    },
    {
      "slug": "minimax-m2-5",
      "date": "2026-02-12",
      "model": "MiniMax M2.5",
      "lab": "MiniMax",
      "lab_slug": "minimax",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "MiniMax Model License",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "MiniMax M2.5 is a language model trained with reinforcement learning across hundreds of thousands of real-world environments, targeting coding, agentic tool use and search, and office work, and completing agentic tasks faster than M2.1.",
      "highlights": [
        "Offered as M2.5 and M2.5-Lightning (highspeed), identical in capability but differing in speed",
        "M2.5-Lightning runs at a steady 100 tokens per second",
        "Weights on Hugging Face under the MiniMax Model License dated February 13, 2026",
        "Trained with the CISPO algorithm plus a process reward mechanism"
      ],
      "source_title": "MiniMax M2.5: Built for Real-World Productivity.",
      "source_url": "https://www.minimax.io/blog/minimax-m25",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/02#minimax-m2-5"
    },
    {
      "slug": "seedance-2-0",
      "date": "2026-02-12",
      "model": "Seedance 2.0",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "video",
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Seedance 2.0 is ByteDance Seed's video creation model built on a unified multimodal audio-video joint generation architecture, accepting text, image, audio, and video inputs with broad reference and editing capabilities.",
      "highlights": [
        "Outputs up to 15 seconds of multi-shot video with dual-channel audio",
        "Supports video editing and extension guided by multimodal references",
        "Available on Jimeng AI, Doubao, and Volcano Ark; BytePlus offers it as Dreamina Seedance 2.0"
      ],
      "source_title": "Seedance 2.0 Official Launch",
      "source_url": "https://seed.bytedance.com/en/blog/seedance-2-0-official-launch",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/02#seedance-2-0"
    },
    {
      "slug": "eleven-v3-conversational",
      "date": "2026-02-10",
      "model": "Eleven v3 Conversational",
      "lab": "ElevenLabs",
      "lab_slug": "elevenlabs",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Enterprise"
      ],
      "summary": "Context-aware text-to-speech model built on Eleven v3 and optimized for real-time dialogue, powering Expressive Mode for ElevenAgents with speech that reflects intent and emotion.",
      "highlights": [
        "Model ID eleven_v3_conversational",
        "Paired with a new turn-taking system using Scribe v2 Realtime",
        "Supports 90+ languages in Expressive Mode"
      ],
      "source_title": "Introducing Expressive Mode",
      "source_url": "https://elevenlabs.io/blog/introducing-expressive-mode",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/02#eleven-v3-conversational"
    },
    {
      "slug": "qwen-image-2-0",
      "date": "2026-02-10",
      "model": "Qwen-Image-2.0",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Qwen-Image-2.0 is Qwen's next-generation foundational image generation model, unifying image generation and editing in a single mode. It targets professional typography such as infographics, slides, posters, and comics.",
      "highlights": [
        "Supports 1K-token instructions",
        "Native 2K resolution",
        "Smaller model size with faster inference than its predecessor; available on Qwen Chat"
      ],
      "source_title": "Qwen-Image-2.0: Professional infographics, exquisite photorealism",
      "source_url": "https://qwen.ai/blog?id=qwen-image-2.0",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/02#qwen-image-2-0"
    },
    {
      "slug": "claude-opus-4-6",
      "date": "2026-02-05",
      "model": "Claude Opus 4.6",
      "lab": "Anthropic",
      "lab_slug": "anthropic",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Cloud platforms"
      ],
      "summary": "Claude Opus 4.6 is an upgrade to Claude Opus 4.5 that plans more carefully, sustains agentic tasks longer, and works more reliably in large codebases. It introduced adaptive thinking and was the first Opus-class model with 1M-token context.",
      "highlights": [
        "API model ID claude-opus-4-6",
        "1M-token context window launched in beta; out of beta on March 13, 2026",
        "128K max output tokens",
        "Compaction API launched in beta alongside the model"
      ],
      "source_title": "Introducing Claude Opus 4.6",
      "source_url": "https://www.anthropic.com/news/claude-opus-4-6",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/02#claude-opus-4-6"
    },
    {
      "slug": "gpt-5-3-codex",
      "date": "2026-02-05",
      "model": "GPT-5.3-Codex",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "coding",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 400000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "GPT-5.3-Codex is an agentic coding model that combines GPT-5.2-Codex coding performance with GPT-5.2 reasoning and professional knowledge in one model, running 25% faster and steerable while it works.",
      "highlights": [
        "Available with paid ChatGPT plans in the Codex app, CLI, IDE extension, and web",
        "Released to the Responses API as gpt-5.3-codex on February 24, 2026",
        "First OpenAI model classified as High capability for cybersecurity under its Preparedness Framework",
        "400,000-token context window per the API model page"
      ],
      "source_title": "Introducing GPT-5.3-Codex | OpenAI",
      "source_url": "https://openai.com/index/introducing-gpt-5-3-codex/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/02#gpt-5-3-codex"
    },
    {
      "slug": "longcat-flash-lite",
      "date": "2026-02-05",
      "model": "LongCat-Flash-Lite",
      "lab": "Meituan",
      "lab_slug": "meituan",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "MoE: 68.5B total, about 3B active",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "LongCat-Flash-Lite is a non-thinking mixture-of-experts model that uses an N-gram embedding table to improve parameter efficiency and inference speed, with a focus on tool use and coding.",
      "highlights": [
        "68.5B total parameters with about 3B activated per token",
        "256K context length through YaRN",
        "N-gram embedding table reduces I/O bottlenecks in MoE layers",
        "Weights on Hugging Face under MIT"
      ],
      "source_title": "Change Log | LongCat API Docs",
      "source_url": "https://longcat.chat/platform/docs/change-log",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/02#longcat-flash-lite"
    },
    {
      "slug": "voxtral-mini-transcribe-2",
      "date": "2026-02-04",
      "model": "Voxtral Mini Transcribe V2",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Batch speech-to-text model in the Voxtral Transcribe 2 family that adds speaker diarization, context biasing and word-level timestamps across 13 languages.",
      "highlights": [
        "API model ID voxtral-mini-2602",
        "Launched with an audio playground in Mistral Studio",
        "Deprecated October 31, 2026 in favor of Voxtral Mini Transcribe 3"
      ],
      "source_title": "Voxtral transcribes at the speed of sound. | Mistral AI",
      "source_url": "https://mistral.ai/news/voxtral-transcribe-2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/02#voxtral-mini-transcribe-2"
    },
    {
      "slug": "voxtral-realtime",
      "date": "2026-02-04",
      "model": "Voxtral Realtime",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "4B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Open-weight live transcription model with a streaming architecture that transcribes audio as it arrives, with delay configurable down to under 200 milliseconds.",
      "highlights": [
        "Released as Voxtral-Mini-4B-Realtime-2602 on Hugging Face",
        "Supports 13 languages",
        "API model ID voxtral-mini-transcribe-realtime-2602"
      ],
      "source_title": "Voxtral transcribes at the speed of sound. | Mistral AI",
      "source_url": "https://mistral.ai/news/voxtral-transcribe-2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/02#voxtral-realtime"
    },
    {
      "slug": "glm-ocr",
      "date": "2026-02-03",
      "model": "GLM-OCR",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "0.9B",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "GLM-OCR is a compact multimodal OCR model for complex document understanding, built on the GLM-V encoder-decoder architecture with a CogViT visual encoder and a GLM-0.5B decoder. It is designed for fast, high-concurrency deployment.",
      "highlights": [
        "Only 0.9B parameters, deployable with vLLM, SGLang, and Ollama",
        "Uses multi-token prediction loss and full-task reinforcement learning",
        "Handles handwriting, tables, seals, and vertical text"
      ],
      "source_title": "Model and product release records (BigModel docs)",
      "source_url": "https://docs.bigmodel.cn/cn/update/new-releases",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/02#glm-ocr"
    },
    {
      "slug": "qwen3-coder-next",
      "date": "2026-02-03",
      "model": "Qwen3-Coder-Next",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 262144,
      "parameters": "MoE: 80B total, 3B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Qwen3-Coder-Next is an open-weights model designed for coding agents and local development, built on Qwen3-Next-80B-A3B-Base. It was agentically trained at scale with executable task synthesis, environment interaction, and reinforcement learning.",
      "highlights": [
        "Hybrid attention and mixture-of-experts architecture from Qwen3-Next",
        "262,144-token native context",
        "Weights under the Apache 2.0 license"
      ],
      "source_title": "Qwen3-Coder-Next: Pushing Small Hybrid Models on Agentic Coding",
      "source_url": "https://qwen.ai/blog?id=qwen3-coder-next",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/02#qwen3-coder-next"
    },
    {
      "slug": "step-3-5-flash",
      "date": "2026-02-02",
      "model": "Step 3.5 Flash",
      "lab": "StepFun",
      "lab_slug": "stepfun",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "MoE: 196B total, 11B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "Step 3.5 Flash is StepFun's open-weight sparse mixture-of-experts reasoning model built for fast agentic and coding work, using three-way multi-token prediction and hybrid sliding-window attention for efficient long-context inference.",
      "highlights": [
        "196B total parameters with about 11B activated per token",
        "256K context window using a 3:1 sliding-window to full-attention layer ratio",
        "3-way multi-token prediction (MTP-3) for faster decoding",
        "Weights on Hugging Face and ModelScope under Apache-2.0; pre-training and mid-training checkpoints followed on March 4, 2026"
      ],
      "source_title": "阶跃星辰 Step 3.5 Flash 上线！更快更强更稳的 Agent 大脑，开源！",
      "source_url": "https://www.stepfun.com/news/step-3-5-flash-launch",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/02#step-3-5-flash"
    },
    {
      "slug": "kling-3-0",
      "date": "2026-01-31",
      "model": "Kling 3.0",
      "lab": "Kuaishou",
      "lab_slug": "kuaishou",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video",
        "audio"
      ],
      "output_modalities": [
        "video",
        "image",
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Kling 3.0 is Kuaishou's model series comprising Video 3.0, Video 3.0 Omni, Image 3.0, and Image 3.0 Omni, unifying generation and editing with native audio, longer clips, and stronger character consistency.",
      "highlights": [
        "Video generation up to 15 seconds with native multilingual audio",
        "Video 3.0 Omni adds reference-video based character and voice consistency",
        "Image 3.0 supports direct 2K and 4K output",
        "Kuaishou's press release announcing the series followed on February 5, 2026"
      ],
      "source_title": "Kling AI Release Notes & Version History | Kling AI",
      "source_url": "https://kling.ai/release-note/release-history",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#kling-3-0"
    },
    {
      "slug": "genie-3",
      "date": "2026-01-29",
      "model": "Genie 3 (Project Genie)",
      "lab": "Google",
      "lab_slug": "google",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Genie 3 is Google DeepMind's world model that generates explorable, interactive environments in real time. First shown as a limited research preview, it became usable through the Project Genie prototype for Google AI Ultra subscribers.",
      "highlights": [
        "Announced August 5, 2025 as a limited research preview",
        "Project Genie opened to U.S. Google AI Ultra subscribers aged 18+ on January 29, 2026",
        "Generations are limited to 60 seconds in Project Genie",
        "Generates worlds navigable in real time at 720p and 24 frames per second"
      ],
      "source_title": "Project Genie: Experimenting with infinite, interactive worlds",
      "source_url": "https://blog.google/innovation-and-ai/models-and-research/google-deepmind/project-genie/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#genie-3"
    },
    {
      "slug": "paddleocr-vl-1-5",
      "date": "2026-01-29",
      "model": "PaddleOCR-VL-1.5",
      "lab": "Baidu",
      "lab_slug": "baidu",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "image",
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "0.9B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "PaddleOCR-VL-1.5 updates Baidu's 0.9B document-parsing vision-language model for in-the-wild robustness, handling skewed, warped, and photographed pages, and adds text spotting and seal recognition.",
      "highlights": [
        "Supports irregular-shaped (polygon) localization for distorted documents",
        "Adds automatic cross-page table merging and cross-page heading recognition",
        "Language coverage adds Tibetan and Bengali"
      ],
      "source_title": "PaddleOCR-VL-1.5: Towards a Multi-Task 0.9B VLM for Robust In-the-Wild Document Parsing",
      "source_url": "https://ernie.baidu.com/research/posts/paddleocr-vl-1.5/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#paddleocr-vl-1-5"
    },
    {
      "slug": "qwen3-asr",
      "date": "2026-01-29",
      "model": "Qwen3-ASR",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "0.6B, 1.7B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Qwen3-ASR is an open-weights speech recognition family of Qwen3-ASR-1.7B and Qwen3-ASR-0.6B, released with the non-autoregressive Qwen3-ForcedAligner-0.6B. The models build on the audio understanding of Qwen3-Omni.",
      "highlights": [
        "Language identification and recognition for 52 languages and accents",
        "Qwen3-ForcedAligner-0.6B performs speech forced alignment",
        "Weights under the Apache 2.0 license"
      ],
      "source_title": "Qwen3-ASR & Qwen3-ForcedAligner is Now Open Sourced",
      "source_url": "https://qwen.ai/blog?id=qwen3asr",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#qwen3-asr"
    },
    {
      "slug": "grok-imagine-video",
      "date": "2026-01-28",
      "model": "Grok Imagine Video (Imagine API)",
      "lab": "xAI",
      "lab_slug": "xai",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "video",
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "The Grok Imagine API brought xAI's video-audio generation and video editing models to developers, supporting image-to-video, text-to-video, and edits such as restyling scenes, adding or removing objects, and controlling motion.",
      "highlights": [
        "API model ID: grok-imagine-video (text, image, or video input per docs)",
        "Generates video with native audio",
        "Docs release notes list a revamped Image Generation API the same day",
        "Offered on partner platforms alongside the xAI API"
      ],
      "source_title": "Grok Imagine API | SpaceXSI",
      "source_url": "https://x.ai/news/grok-imagine-api",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#grok-imagine-video"
    },
    {
      "slug": "deepseek-ocr-2",
      "date": "2026-01-27",
      "model": "DeepSeek-OCR 2",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "DeepSeek-OCR 2 is DeepSeek's second document OCR vision-language model, introducing a visual encoding approach the lab calls Visual Causal Flow. It follows DeepSeek-OCR and is released with vLLM and Transformers inference code.",
      "highlights": [
        "Subtitle: Visual Causal Flow, aimed at more human-like visual encoding",
        "Weights on Hugging Face under the Apache 2.0 license",
        "Announced in the DeepSeek-OCR repository on January 27, 2026"
      ],
      "source_title": "DeepSeek-OCR: Contexts Optical Compression (GitHub)",
      "source_url": "https://github.com/deepseek-ai/DeepSeek-OCR",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#deepseek-ocr-2"
    },
    {
      "slug": "kimi-k2-5",
      "date": "2026-01-27",
      "model": "Kimi K2.5",
      "lab": "Moonshot AI",
      "lab_slug": "moonshot-ai",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "MoE: 1T total, 32B active",
      "model_license": "Modified MIT License",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "Kimi K2.5 builds on Kimi K2 with continued pre-training on about 15T mixed visual and text tokens, making it a native multimodal model. It introduced a self-directed agent swarm that can coordinate up to 100 sub-agents.",
      "highlights": [
        "Agent swarm runs parallel workflows across up to 1,500 tool calls",
        "Kimi.com and the Kimi App offer Instant, Thinking, Agent, and Agent Swarm (beta) modes",
        "Weights and code under the Modified MIT License"
      ],
      "source_title": "Kimi K2.5: Visual Agentic Intelligence",
      "source_url": "https://www.kimi.com/en/blog/kimi-k2-5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#kimi-k2-5"
    },
    {
      "slug": "minimax-m2-her",
      "date": "2026-01-27",
      "model": "MiniMax M2-her",
      "lab": "MiniMax",
      "lab_slug": "minimax",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "MiniMax-M2-her is a role-play language model for multi-turn character dialogue, built from three years of role-play optimization in MiniMax's Talkie and Xingye apps and offered through the API as a chat model.",
      "highlights": [
        "Designed to keep characters consistent with established world settings and lore",
        "Uses narration to bridge time or scene changes instead of forcing impossible dialogue",
        "Listed in the MiniMax API docs as the M2-her chat model"
      ],
      "source_title": "A Deep Dive into the MiniMax-M2-her",
      "source_url": "https://www.minimax.io/news/a-deep-dive-into-the-minimax-m2-her-2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#minimax-m2-her"
    },
    {
      "slug": "sera",
      "date": "2026-01-27",
      "model": "SERA",
      "lab": "Ai2",
      "lab_slug": "ai2",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "8B to 32B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "First release in Ai2's Open Coding Agents family, coding models built on Qwen 3 with a training method for adapting agents to any repository.",
      "highlights": [
        "Trained up to 32K context length",
        "SERA-14B added on February 3, 2026",
        "Released under Apache 2.0"
      ],
      "source_title": "Open Coding Agents: Fast, accessible coding agents that adapt to any repo | Ai2",
      "source_url": "https://allenai.org/blog/open-coding-agents",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#sera"
    },
    {
      "slug": "hunyuanimage-3-0-instruct",
      "date": "2026-01-26",
      "model": "HunyuanImage 3.0-Instruct",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Tencent Hunyuan Community License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "HunyuanImage 3.0-Instruct adds instruction reasoning to HunyuanImage 3.0 for prompt enhancement, along with image-to-image generation for editing and multi-image fusion. A distilled checkpoint for faster sampling was released alongside it.",
      "highlights": [
        "HunyuanImage-3.0-Instruct-Distil recommends 8-step sampling",
        "Weights on Hugging Face"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/HunyuanImage-3.0: HunyuanImage-3.0: A Powerful Native Multimodal Model for Image Generation",
      "source_url": "https://github.com/Tencent-Hunyuan/HunyuanImage-3.0",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#hunyuanimage-3-0-instruct"
    },
    {
      "slug": "qwen3-max-thinking",
      "date": "2026-01-26",
      "model": "Qwen3-Max-Thinking",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Qwen3-Max-Thinking is Qwen's flagship reasoning model, built by scaling model parameters and reinforcement learning compute on Qwen3-Max. It adds adaptive tool use for on-demand retrieval and code interpreter calls.",
      "highlights": [
        "Available in Qwen Chat and through the Alibaba Cloud API",
        "Uses test-time scaling techniques to improve reasoning",
        "Improves factual knowledge, instruction following, and agent capabilities"
      ],
      "source_title": "Pushing Qwen3-Max-Thinking Beyond its Limits",
      "source_url": "https://qwen.ai/blog?id=qwen3-max-thinking",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#qwen3-max-thinking"
    },
    {
      "slug": "luma-ray3-14",
      "date": "2026-01-26",
      "model": "Ray3.14",
      "lab": "Luma AI",
      "lab_slug": "luma-ai",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Ray3 update adding native 1080p generation across core Dream Machine workflows, four times faster generation at 720p than Ray3, better prompt adherence and more consistent Modify Video output.",
      "highlights": [
        "Native 1080p generation",
        "Modify Video up to 18 seconds",
        "Character references not supported at launch"
      ],
      "source_title": "Ray3.14 is here: Native 1080p, 3x cheaper and 4x faster.",
      "source_url": "https://lumalabs.ai/news/ray3_14",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#luma-ray3-14"
    },
    {
      "slug": "minimax-speech-2-8",
      "date": "2026-01-23",
      "model": "MiniMax Speech 2.8",
      "lab": "MiniMax",
      "lab_slug": "minimax",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "MiniMax Speech 2.8 is a text-to-speech model that adds native sound tags for breaths, laughter, and filler words, along with higher-fidelity voice cloning and cleaner, studio-grade audio.",
      "highlights": [
        "Offered as speech-2.8-hd and speech-2.8-turbo",
        "Voice cloning from about 10 seconds of audio",
        "Improved cross-language rendering starting with the Mandarin-Japanese pair"
      ],
      "source_title": "MiniMax Speech 2.8: Breathing life into AI voice",
      "source_url": "https://www.minimax.io/news/minimax-speech-28",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#minimax-speech-2-8"
    },
    {
      "slug": "qwen3-tts",
      "date": "2026-01-22",
      "model": "Qwen3-TTS",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": "0.6B, 1.7B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Qwen3-TTS is Qwen's open-weights speech generation family supporting voice cloning, voice design, and natural-language voice control. It is built on the Qwen3-TTS-Tokenizer-12Hz multi-codebook speech encoder.",
      "highlights": [
        "Released in 1.7B and 0.6B sizes",
        "Supports 10 major languages plus various dialects",
        "Weights under the Apache 2.0 license"
      ],
      "source_title": "Qwen3-TTS Family is Now Open Sourced: Voice Design, Clone, and Generation!",
      "source_url": "https://qwen.ai/blog?id=qwen3tts-0115",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#qwen3-tts"
    },
    {
      "slug": "lfm2-5-1-2b-thinking",
      "date": "2026-01-20",
      "model": "LFM2.5-1.2B-Thinking",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32768,
      "parameters": "1.17B",
      "model_license": "LFM Open License v1.0",
      "available_at_launch": [
        "Consumer app",
        "Open weights"
      ],
      "summary": "A 1.2B-parameter reasoning model in the LFM2.5 family that generates thinking traces before answering and runs on-device within 900 MB of memory on a phone.",
      "highlights": [
        "Available on Hugging Face, LEAP, and Liquid Playground",
        "New launch partners: Qualcomm Technologies, Ollama, FastFlowLM, and Cactus Compute",
        "32,768-token context window per model card"
      ],
      "source_title": "LFM2.5-1.2B-Thinking: On-Device Reasoning Under 1GB",
      "source_url": "https://www.liquid.ai/blog/lfm2-5-1-2b-thinking-on-device-reasoning-under-1gb",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#lfm2-5-1-2b-thinking"
    },
    {
      "slug": "glm-4-7-flash",
      "date": "2026-01-19",
      "model": "GLM-4.7-Flash",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "MoE: 30B total, 3B active",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "GLM-4.7-Flash is a lightweight 30B-A3B mixture-of-experts version of GLM-4.7 for efficient local and API deployment. Z.ai optimized it for high-frequency use cases on its API platform.",
      "highlights": [
        "Lightweight version of GLM-4.7 offered on the BigModel and Z.ai APIs",
        "Aimed at coding, writing, translation, and long-text tasks",
        "Weights on Hugging Face under the MIT license"
      ],
      "source_title": "Model and product release records (BigModel docs)",
      "source_url": "https://docs.bigmodel.cn/cn/update/new-releases",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#glm-4-7-flash"
    },
    {
      "slug": "minimax-music-2-5",
      "date": "2026-01-16",
      "model": "MiniMax Music 2.5",
      "lab": "MiniMax",
      "lab_slug": "minimax",
      "type": "music",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "MiniMax Music 2.5 adds paragraph-level control over song structure and higher-fidelity sound, supporting 14 structural section tags such as Intro, Bridge, Interlude, Build-up, and Hook.",
      "highlights": [
        "API release notes list it on January 16, 2026; the announcement post is dated January 28, 2026",
        "Supports more than 100 instruments",
        "Lyrics can carry structural tags, instrument names, and prompts"
      ],
      "source_title": "MiniMax Music 2.5: Breakthrough Across All Dimensions, Direct the Detail. Define the Real",
      "source_url": "https://www.minimax.io/news/minimax-music-25",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#minimax-music-2-5"
    },
    {
      "slug": "flux-2-klein-4b",
      "date": "2026-01-15",
      "model": "FLUX.2 [klein] 4B",
      "lab": "Black Forest Labs",
      "lab_slug": "black-forest-labs",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "4B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Compact FLUX.2 model family member that unifies text-to-image, editing and multi-reference generation with sub-second inference, fully open under Apache 2.0 and able to run in about 13GB of VRAM.",
      "highlights": [
        "Distilled and undistilled Base 4B variants",
        "FP8 and NVFP4 builds developed with NVIDIA",
        "Also served through the BFL API"
      ],
      "source_title": "FLUX.2 [klein]: Towards Interactive Visual Intelligence",
      "source_url": "https://bfl.ai/blog/flux2-klein-towards-interactive-visual-intelligence",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#flux-2-klein-4b"
    },
    {
      "slug": "flux-2-klein-9b",
      "date": "2026-01-15",
      "model": "FLUX.2 [klein] 9B",
      "lab": "Black Forest Labs",
      "lab_slug": "black-forest-labs",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "9B",
      "model_license": "FLUX Non-Commercial License",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Flagship small FLUX.2 model, a 9B flow model with an 8B Qwen3 text embedder step-distilled to four inference steps for sub-second generation, editing and multi-reference work.",
      "highlights": [
        "Undistilled Base 9B variant for fine-tuning and research",
        "Open weights under the FLUX Non-Commercial License",
        "Available in the BFL API"
      ],
      "source_title": "FLUX.2 [klein]: Towards Interactive Visual Intelligence",
      "source_url": "https://bfl.ai/blog/flux2-klein-towards-interactive-visual-intelligence",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#flux-2-klein-9b"
    },
    {
      "slug": "translategemma",
      "date": "2026-01-15",
      "model": "TranslateGemma",
      "lab": "Google",
      "lab_slug": "google",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 2048,
      "parameters": "4B, 12B, 27B",
      "model_license": "Gemma Terms of Use",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "TranslateGemma is a suite of open translation models built on Gemma 3 in 4B, 12B and 27B sizes. It covers 55 languages and keeps Gemma 3's multimodal ability to translate text found inside images.",
      "highlights": [
        "4B targets mobile and edge, 12B consumer laptops, 27B a single H100 GPU or TPU",
        "Evaluated on 55 language pairs, with nearly 500 more pairs in training",
        "Weights on Kaggle and Hugging Face; deployable in Vertex AI"
      ],
      "source_title": "TranslateGemma: A new suite of open translation models",
      "source_url": "https://blog.google/innovation-and-ai/technology/developers-tools/translategemma/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#translategemma"
    },
    {
      "slug": "glm-image",
      "date": "2026-01-14",
      "model": "GLM-Image",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "9B autoregressive generator + 7B diffusion decoder",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "GLM-Image is Z.ai's image generation model that combines an autoregressive generator with a diffusion decoder. Z.ai says it was fully trained on domestic Chinese chips and improves text rendering in knowledge-intensive images.",
      "highlights": [
        "Autoregressive generator initialized from GLM-4-9B-0414",
        "7B single-stream DiT decoder with a Glyph Encoder for accurate in-image text",
        "Weights on Hugging Face under the MIT license"
      ],
      "source_title": "GLM-Image: Auto-regressive for Dense-knowledge and High-fidelity Image Generation",
      "source_url": "https://z.ai/blog/glm-image",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#glm-image"
    },
    {
      "slug": "longcat-flash-thinking-2601",
      "date": "2026-01-14",
      "model": "LongCat-Flash-Thinking-2601",
      "lab": "Meituan",
      "lab_slug": "meituan",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "MoE: 560B total, about 27B active",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "LongCat-Flash-Thinking-2601 is an upgraded LongCat reasoning model that strengthens agentic tool use and search through large-scale multi-environment reinforcement learning and adds a deep thinking mode.",
      "highlights": [
        "Trained across environments with more than 60 tool types",
        "Deep thinking mode combines parallel reasoning with recursive summarization",
        "A LongCat-Flash-Thinking-ZigZag variant with sparse attention was also released",
        "Weights on Hugging Face and GitHub under MIT"
      ],
      "source_title": "Change Log | LongCat API Docs",
      "source_url": "https://longcat.chat/platform/docs/change-log",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#longcat-flash-thinking-2601"
    },
    {
      "slug": "step-audio-r1-1",
      "date": "2026-01-14",
      "model": "Step-Audio R1.1",
      "lab": "StepFun",
      "lab_slug": "stepfun",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "audio",
        "text"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Step-Audio R1.1 (Realtime) is an upgrade to Step-Audio-R1 for interactive spoken dialogue, designed to reason while speaking so that real-time voice conversations keep low latency without giving up reasoning ability.",
      "highlights": [
        "Dual-brain design separates high-level reasoning from speech generation",
        "Grounds reasoning in acoustic representations rather than transcripts alone",
        "Inference code and weights released on Hugging Face and ModelScope"
      ],
      "source_title": "GitHub - stepfun-ai/Step-Audio-R1",
      "source_url": "https://github.com/stepfun-ai/Step-Audio-R1",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#step-audio-r1-1"
    },
    {
      "slug": "medasr",
      "date": "2026-01-13",
      "model": "MedASR",
      "lab": "Google",
      "lab_slug": "google",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "105M",
      "model_license": "Health AI Developer Foundations Terms of Use",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "MedASR is an open automatic speech recognition model from Google Research fine-tuned for medical dictation. Released alongside MedGemma 1.5, it converts clinical speech to text and can produce prompts for MedGemma.",
      "highlights": [
        "Conformer-based model with 105M parameters per its model card",
        "English-only, trained for medical terminology such as radiology dictation",
        "Available on Hugging Face and Vertex AI"
      ],
      "source_title": "Next generation medical image interpretation with MedGemma 1.5 and medical speech to text with MedASR",
      "source_url": "https://research.google/blog/next-generation-medical-image-interpretation-with-medgemma-15-and-medical-speech-to-text-with-medasr/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#medasr"
    },
    {
      "slug": "medgemma-1-5",
      "date": "2026-01-13",
      "model": "MedGemma 1.5",
      "lab": "Google",
      "lab_slug": "google",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "4B",
      "model_license": "Health AI Developer Foundations Terms of Use",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "MedGemma 1.5 is the next-generation 4B MedGemma model for medical image and text understanding. It adds support for 3D CT and MRI volumes, whole-slide histopathology, longitudinal chest X-rays and medical document extraction.",
      "highlights": [
        "Accepts multiple slices or patches of high-dimensional scans alongside a text prompt",
        "Adds anatomical localization in chest X-rays",
        "Retains MedGemma 1's 2D image and text abilities",
        "Available on Hugging Face and Vertex AI; free for research and commercial use under HAI-DEF terms"
      ],
      "source_title": "Next generation medical image interpretation with MedGemma 1.5 and medical speech to text with MedASR",
      "source_url": "https://research.google/blog/next-generation-medical-image-interpretation-with-medgemma-15-and-medical-speech-to-text-with-medasr/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#medgemma-1-5"
    },
    {
      "slug": "step3-vl-10b",
      "date": "2026-01-13",
      "model": "Step3-VL-10B",
      "lab": "StepFun",
      "lab_slug": "stepfun",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "10B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Step3-VL-10B is a compact open-weight multimodal foundation model from StepFun that targets visual perception, complex reasoning, and human-centric alignment at a 10B-parameter scale.",
      "highlights": [
        "10B-parameter vision-language model with a base checkpoint also released",
        "Optional PaCoRe parallel reasoning mode aggregates 16 parallel rollouts",
        "FP8 weights and official vLLM support followed the initial release"
      ],
      "source_title": "stepfun-ai/Step3-VL-10B · Hugging Face",
      "source_url": "https://huggingface.co/stepfun-ai/Step3-VL-10B",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#step3-vl-10b"
    },
    {
      "slug": "niji-7",
      "date": "2026-01-09",
      "model": "Niji 7",
      "lab": "Midjourney",
      "lab_slug": "midjourney",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Niji model tuned for Eastern and anime aesthetics with a major boost in coherency, clearer fine details, closer prompt following and improved text rendering and style reference performance.",
      "highlights": [
        "Developed with Spellbrush",
        "More literal prompt interpretation than earlier Niji models",
        "Cleaner, flatter look highlighting line work"
      ],
      "source_title": "Version - Midjourney",
      "source_url": "https://docs.midjourney.com/hc/en-us/articles/32199405667853-Version",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#niji-7"
    },
    {
      "slug": "elevenlabs-scribe-v2",
      "date": "2026-01-09",
      "model": "Scribe v2",
      "lab": "ElevenLabs",
      "lab_slug": "elevenlabs",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Batch speech-to-text model for transcription, subtitling and captioning at scale, improving on Scribe v1 for long-form audio, pauses and tone changes and adding keyterm prompting.",
      "highlights": [
        "Keyterm prompting and built-in entity detection",
        "Automatic multi-language transcription",
        "Entity redaction and No Verbatim mode added April 2026"
      ],
      "source_title": "Introducing Scribe v2",
      "source_url": "https://elevenlabs.io/blog/introducing-scribe-v2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#elevenlabs-scribe-v2"
    },
    {
      "slug": "jamba2",
      "date": "2026-01-08",
      "model": "Jamba2",
      "lab": "AI21 Labs",
      "lab_slug": "ai21",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "3B dense; Mini MoE 52B total, 12B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Second-generation Jamba family of open models built for enterprise reliability, released as a dense 3B model for devices and a mixture-of-experts Mini model.",
      "highlights": [
        "256K-token context window",
        "Mini aims at precise question answering without a reasoning model's overhead",
        "Released under Apache 2.0"
      ],
      "source_title": "Introducing Jamba2: The open source model family for enterprise reliability and efficiency",
      "source_url": "https://www.ai21.com/blog/introducing-jamba2/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#jamba2"
    },
    {
      "slug": "qwen3-vl-embedding",
      "date": "2026-01-08",
      "model": "Qwen3-VL-Embedding and Qwen3-VL-Reranker",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "embedding",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32000,
      "parameters": "2B, 8B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Qwen3-VL-Embedding and Qwen3-VL-Reranker are open-weights multimodal retrieval models built on Qwen3-VL. They handle text, images, screenshots, and video in one framework for embedding and reranking.",
      "highlights": [
        "Supports more than 30 languages",
        "Flexible embedding dimensions and customizable instructions",
        "Released in 2B and 8B sizes under the Apache 2.0 license"
      ],
      "source_title": "Qwen3-VL-Embedding and Qwen3-VL-Reranker",
      "source_url": "https://qwen.ai/blog?id=qwen3-vl-embedding",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#qwen3-vl-embedding"
    },
    {
      "slug": "lfm2-2-6b-transcript",
      "date": "2026-01-07",
      "model": "LFM2-2.6B-Transcript",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "2.6B",
      "model_license": "LFM Open License v1.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "A Liquid Nano fine-tuned from LFM2-2.6B for summarizing 30 to 60 minute meeting transcripts on-device, developed with AMD and showcased at CES running locally across Ryzen AI CPUs, GPUs, and NPUs.",
      "highlights": [
        "Outputs structured summaries with key points, decisions, and attributed action items",
        "Under 3 GB RAM usage reported for the AMD deployment",
        "Available for download on Hugging Face and LEAP"
      ],
      "source_title": "The Future of Meeting Summarization: Local, Fast, Private, and Fully Secure",
      "source_url": "https://www.liquid.ai/blog/the-future-of-meeting-summarization-local-fast-private-and-fully-secure",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#lfm2-2-6b-transcript"
    },
    {
      "slug": "nouscoder-14b",
      "date": "2026-01-06",
      "model": "NousCoder-14B",
      "lab": "Nous Research",
      "lab_slug": "nous-research",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "14B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Competitive olympiad programming model from Nous Research, post-trained from Qwen3-14B with reinforcement learning to solve contest-style coding problems and released with open weights under Apache 2.0.",
      "highlights": [
        "Post-trained on Qwen3-14B",
        "Training context extended from 32K to 40K tokens during RL",
        "Released under Apache 2.0"
      ],
      "source_title": "NousCoder-14B: A Competitive Olympiad Programming Model | Nous Research",
      "source_url": "https://nousresearch.com/nouscoder-14b-a-competitive-olympiad-programming-model",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#nouscoder-14b"
    },
    {
      "slug": "lfm2-5-1-2b",
      "date": "2026-01-05",
      "model": "LFM2.5-1.2B (Base, Instruct, JP)",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32768,
      "parameters": "1.17B",
      "model_license": "LFM Open License v1.0",
      "available_at_launch": [
        "Consumer app",
        "Open weights"
      ],
      "summary": "The first LFM2.5 release, a 1.2B text family on the LFM2 architecture with pretraining extended from 10T to 28T tokens and expanded reinforcement learning, shipped as Base, Instruct, and Japanese-optimized checkpoints.",
      "highlights": [
        "Checkpoints: LFM2.5-1.2B-Base, LFM2.5-1.2B-Instruct, and LFM2.5-1.2B-JP",
        "Announced alongside LFM2.5-VL-1.6B and LFM2.5-Audio-1.5B",
        "Open weights on Hugging Face and LEAP; Instruct also on Liquid Playground",
        "AMD and Nexa AI provide NPU-optimized versions"
      ],
      "source_title": "Introducing LFM2.5: The Next Generation of On-Device AI",
      "source_url": "https://www.liquid.ai/blog/introducing-lfm2-5-the-next-generation-of-on-device-ai",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#lfm2-5-1-2b"
    },
    {
      "slug": "lfm2-5-audio-1-5b",
      "date": "2026-01-05",
      "model": "LFM2.5-Audio-1.5B",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": 32768,
      "parameters": "1.5B",
      "model_license": "LFM Open License v1.0",
      "available_at_launch": [
        "Consumer app",
        "Open weights"
      ],
      "summary": "A native audio-language model in the LFM2.5 family that accepts and produces both speech and text, introducing a custom LFM-based audio detokenizer that Liquid AI says is 8x faster than LFM2's Mimi detokenizer.",
      "highlights": [
        "llama.cpp-compatible GGUF checkpoints for CPU inference",
        "Improved ASR and TTS quality over LFM2-Audio-1.5B per Liquid AI",
        "Available on Hugging Face, LEAP, and Liquid Playground"
      ],
      "source_title": "Introducing LFM2.5: The Next Generation of On-Device AI",
      "source_url": "https://www.liquid.ai/blog/introducing-lfm2-5-the-next-generation-of-on-device-ai",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#lfm2-5-audio-1-5b"
    },
    {
      "slug": "lfm2-5-vl-1-6b",
      "date": "2026-01-05",
      "model": "LFM2.5-VL-1.6B",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32768,
      "parameters": "1.6B",
      "model_license": "LFM Open License v1.0",
      "available_at_launch": [
        "Consumer app",
        "Open weights"
      ],
      "summary": "A refreshed 1.6B vision-language model on the updated LFM2.5 base backbone, released with the LFM2.5 family and tuned for multi-image comprehension, multilingual visual understanding, and instruction following.",
      "highlights": [
        "Handles prompts in Arabic, Chinese, French, German, Japanese, Korean, and Spanish",
        "Succeeds LFM2-VL-1.6B",
        "Available on Hugging Face, LEAP, and Liquid Playground"
      ],
      "source_title": "Introducing LFM2.5: The Next Generation of On-Device AI",
      "source_url": "https://www.liquid.ai/blog/introducing-lfm2-5-the-next-generation-of-on-device-ai",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2026/01#lfm2-5-vl-1-6b"
    },
    {
      "slug": "qwen-image-2512",
      "date": "2025-12-31",
      "model": "Qwen-Image-2512",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "20B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Consumer app",
        "Open weights"
      ],
      "summary": "Qwen-Image-2512 is the December 2025 update of Qwen-Image's text-to-image foundation model. Qwen says it reduces the AI-generated look, improves realism for human subjects, and renders finer natural detail.",
      "highlights": [
        "Weights on Hugging Face and ModelScope on release day",
        "Available to try on Qwen Chat",
        "Released under the Apache 2.0 license"
      ],
      "source_title": "Qwen-Image-2512: Finer Details, Greater Realism",
      "source_url": "https://qwen.ai/blog?id=qwen-image-2512",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#qwen-image-2512"
    },
    {
      "slug": "hy-motion-1-0",
      "date": "2025-12-30",
      "model": "HY-Motion 1.0",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "3d"
      ],
      "context_window": null,
      "parameters": "1.0B",
      "model_license": "Tencent HY-Motion 1.0 Community License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "HY-Motion 1.0 is an open series of text-to-3D human motion generation models based on a Diffusion Transformer and flow matching, producing skeleton-based 3D character animations from text prompts.",
      "highlights": [
        "Standard model with 1.0B parameters and a Lite model with 0.46B",
        "Tencent describes it as the first DiT-based text-to-motion model scaled to a billion parameters"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/HY-Motion-1.0: HY-Motion model for 3D human motion or 3D character animation generation.",
      "source_url": "https://github.com/Tencent-Hunyuan/HY-Motion-1.0",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#hy-motion-1-0"
    },
    {
      "slug": "hy-mt1-5",
      "date": "2025-12-30",
      "model": "HY-MT1.5",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "1.8B, 7B",
      "model_license": "Tencent HY Community License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "HY-MT1.5 is Tencent's second open translation model release, with 1.8B and 7B models supporting mutual translation across 33 languages and five ethnic and dialect variations. The 1.8B model targets edge devices and real-time translation.",
      "highlights": [
        "HY-MT1.5-7B is optimized for annotated and mixed-language text over the September 2025 release",
        "FP8, GPTQ-Int4, and GGUF versions published"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/Hy-MT",
      "source_url": "https://github.com/Tencent-Hunyuan/Hy-MT",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#hy-mt1-5"
    },
    {
      "slug": "nextstep-1-1",
      "date": "2025-12-24",
      "model": "NextStep-1.1",
      "lab": "StepFun",
      "lab_slug": "stepfun",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "NextStep-1.1 is an updated version of StepFun's autoregressive text-to-image model that improves output quality through extended training and a flow-based reinforcement learning post-training stage.",
      "highlights": [
        "Builds on NextStep-1's autoregressive generation with continuous image tokens",
        "Adds flow-based reinforcement learning post-training",
        "Checkpoints hosted on Hugging Face under Apache-2.0"
      ],
      "source_title": "GitHub - stepfun-ai/NextStep-1",
      "source_url": "https://github.com/stepfun-ai/NextStep-1",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#nextstep-1-1"
    },
    {
      "slug": "minimax-m2-1",
      "date": "2025-12-23",
      "model": "MiniMax M2.1",
      "lab": "MiniMax",
      "lab_slug": "minimax",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Modified-MIT",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "MiniMax M2.1 updates M2 for real-world complex tasks, with stronger programming in Rust, Java, Go, C++, Kotlin, Objective-C, TypeScript, and JavaScript, plus improved web and mobile app development and office scenarios.",
      "highlights": [
        "Released on the API as MiniMax-M2.1 and M2.1-highspeed",
        "Weights open-sourced on Hugging Face under a modified MIT license",
        "Powers the MiniMax Agent product"
      ],
      "source_title": "MiniMax M2.1: Significantly Enhanced Multi-Language Programming, Built for Real-World Complex Tasks",
      "source_url": "https://www.minimax.io/blog/minimax-m21",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#minimax-m2-1"
    },
    {
      "slug": "qwen-image-edit-2511",
      "date": "2025-12-23",
      "model": "Qwen-Image-Edit-2511",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "20B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Consumer app",
        "Open weights"
      ],
      "summary": "Qwen-Image-Edit-2511 is an enhanced version of Qwen-Image-Edit-2509 with better consistency. It mitigates image drift and improves character consistency in edits.",
      "highlights": [
        "Available through the Image Editing feature on Qwen Chat",
        "Weights on Hugging Face and ModelScope",
        "Released under the Apache 2.0 license"
      ],
      "source_title": "Qwen-Image-Edit-2511: Improve Consistency",
      "source_url": "https://qwen.ai/blog?id=qwen-image-edit-2511",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#qwen-image-edit-2511"
    },
    {
      "slug": "glm-4-7",
      "date": "2025-12-22",
      "model": "GLM-4.7",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 200000,
      "parameters": "MoE: 355B total, 32B active",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "GLM-4.7 is Z.ai's foundation model with improvements in coding, reasoning, and agentic capabilities over GLM-4.6. It supports thinking before acting in agent frameworks such as Claude Code, Kilo Code, Cline, and Roo Code.",
      "highlights": [
        "Gains in multilingual agentic coding and terminal-based tasks",
        "Cleaner webpage and slide generation",
        "Weights on Hugging Face and ModelScope under the MIT license"
      ],
      "source_title": "GLM-4.7: Advancing the Coding Capability",
      "source_url": "https://z.ai/blog/glm-4.7",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#glm-4-7"
    },
    {
      "slug": "longcat-flash-chat-2512",
      "date": "2025-12-22",
      "model": "LongCat-Flash-Chat (December 2025 update)",
      "lab": "Meituan",
      "lab_slug": "meituan",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Meituan upgraded the LongCat-Flash-Chat model on its API with stronger programming performance, a doubled context length, broader multilingual support, and more robust agent behavior, without changing the model name.",
      "highlights": [
        "Context length doubled from 128K to 256K tokens",
        "High-quality support for nine languages, including Spanish, French, and Arabic",
        "Model name and API invocation unchanged"
      ],
      "source_title": "Change Log | LongCat API Docs",
      "source_url": "https://longcat.chat/platform/docs/change-log",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#longcat-flash-chat-2512"
    },
    {
      "slug": "qwen-image-layered",
      "date": "2025-12-19",
      "model": "Qwen-Image-Layered",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Qwen-Image-Layered is a Qwen model that decomposes an image into multiple RGBA layers, so each layer can be edited independently. The layered output supports resizing, repositioning, and recoloring without affecting other content.",
      "highlights": [
        "Weights on Hugging Face and ModelScope",
        "Research paper published December 18, 2025",
        "Released under the Apache 2.0 license"
      ],
      "source_title": "Qwen-Image-Layered: Layered Decomposition for Inherent Editablity",
      "source_url": "https://qwen.ai/blog?id=qwen-image-layered",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#qwen-image-layered"
    },
    {
      "slug": "functiongemma",
      "date": "2025-12-18",
      "model": "FunctionGemma",
      "lab": "Google",
      "lab_slug": "google",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32768,
      "parameters": "270M",
      "model_license": "Gemma Terms of Use",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "FunctionGemma is a version of Gemma 3 270M fine-tuned for function calling, intended as a base for fast, private, on-device agents. It can handle tasks offline or route complex requests to larger models.",
      "highlights": [
        "Small enough to run on NVIDIA Jetson Nano and mobile phones",
        "32K-token input context per the model card",
        "Weights on Hugging Face and Kaggle"
      ],
      "source_title": "FunctionGemma: Bringing bespoke function calling to the edge",
      "source_url": "https://blog.google/technology/developers/functiongemma/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#functiongemma"
    },
    {
      "slug": "gpt-5-2-codex",
      "date": "2025-12-18",
      "model": "GPT-5.2-Codex",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "coding",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 400000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "GPT-5.2-Codex is a version of GPT-5.2 further optimized for agentic coding in Codex, with better long-horizon work through context compaction, stronger handling of large refactors and migrations, and stronger cybersecurity capabilities.",
      "highlights": [
        "Released in all Codex surfaces for paid ChatGPT users; CLI and IDE extension default to gpt-5.2-codex",
        "API model ID gpt-5.2-codex reached the Responses API on January 14, 2026",
        "Improved performance in Windows environments",
        "400,000-token context window per the API model page"
      ],
      "source_title": "Introducing GPT-5.2-Codex | OpenAI",
      "source_url": "https://openai.com/index/introducing-gpt-5-2-codex/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#gpt-5-2-codex"
    },
    {
      "slug": "luma-ray3-modify",
      "date": "2025-12-18",
      "model": "Ray3 Modify",
      "lab": "Luma AI",
      "lab_slug": "luma-ai",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Ray3-based video-to-video model for editing and reimagining footage, adding start and end frame keyframe control and character references while preserving the original performance.",
      "highlights": [
        "First start and end frame control in Luma's video-to-video workflow",
        "Character Reference for consistent subjects",
        "Available in Dream Machine"
      ],
      "source_title": "Introducing Ray3 Modify",
      "source_url": "https://lumalabs.ai/news/ray3-modify",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#luma-ray3-modify"
    },
    {
      "slug": "seed-1-8",
      "date": "2025-12-18",
      "model": "Seed1.8",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Seed1.8 is a generalized agentic model that combines search, coding, and GUI agent capabilities with native vision, accepting text and image input and offering three thinking modes to balance latency against depth.",
      "highlights": [
        "Integrates search, code, and GUI agent abilities in one model",
        "Three thinking modes adjust to task complexity",
        "Reduces the tokens needed to encode images and video",
        "Available on Volcano Engine as Doubao 1.8"
      ],
      "source_title": "Official Release of Seed1.8: A Generalized Agentic Model",
      "source_url": "https://seed.bytedance.com/en/blog/official-release-of-seed1-8-a-generalized-agentic-model",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#seed-1-8"
    },
    {
      "slug": "t5gemma-2",
      "date": "2025-12-18",
      "model": "T5Gemma 2",
      "lab": "Google",
      "lab_slug": "google",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 131072,
      "parameters": "270M-270M, 1B-1B, 4B-4B",
      "model_license": "Gemma Terms of Use",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "T5Gemma 2 is the second generation of Google's encoder-decoder Gemma models, now based on Gemma 3. It adds image understanding, a 128K-token context window and support for over 140 languages.",
      "highlights": [
        "Sizes total about 370M, 1.7B and 7B parameters, excluding the vision encoder",
        "Handles visual question answering and multimodal reasoning",
        "Only pre-trained checkpoints released, no instruction-tuned versions"
      ],
      "source_title": "T5Gemma 2: The next generation of encoder-decoder models",
      "source_url": "https://blog.google/technology/developers/t5gemma-2/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#t5gemma-2"
    },
    {
      "slug": "gemini-3-flash",
      "date": "2025-12-17",
      "model": "Gemini 3 Flash",
      "lab": "Google",
      "lab_slug": "google",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1048576,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms",
        "Consumer app",
        "Enterprise"
      ],
      "summary": "Gemini 3 Flash brings Gemini 3 frontier intelligence to a fast, lower-cost Flash model with upgraded visual and spatial reasoning and agentic coding, rolling out globally in the Gemini app and AI Mode in Search.",
      "highlights": [
        "Preview model gemini-3-flash-preview in the Gemini API, Google Antigravity, Gemini CLI, Android Studio and Vertex AI",
        "Also available in Gemini Enterprise",
        "Input token limit of 1,048,576 per the Gemini API model page"
      ],
      "source_title": "Introducing Gemini 3 Flash: Benchmarks, global availability",
      "source_url": "https://blog.google/products-and-platforms/products/gemini/gemini-3-flash/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#gemini-3-flash"
    },
    {
      "slug": "grok-voice-agent-api",
      "date": "2025-12-17",
      "model": "Grok Voice Agent API",
      "lab": "xAI",
      "lab_slug": "xai",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "The Grok Voice Agent API opened xAI's in-house speech-to-speech stack, which powers Grok Voice in its apps and Tesla vehicles, to developers building multilingual voice agents that call tools and search real-time data.",
      "highlights": [
        "Voice activity detection, tokenizer, and audio models trained in-house from scratch",
        "Average time-to-first-audio under 1 second, per xAI",
        "Compatible with the OpenAI Realtime API specification; LiveKit plugin available",
        "Voices include Ara, Eve, and Leo"
      ],
      "source_title": "Grok Voice Agent API | SpaceXSI",
      "source_url": "https://x.ai/news/grok-voice-agent-api",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#grok-voice-agent-api"
    },
    {
      "slug": "hy-world-1-5-worldplay",
      "date": "2025-12-17",
      "model": "HY-World 1.5 (WorldPlay)",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Tencent HY-WorldPlay Community License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "HY-World 1.5 (WorldPlay) is an open streaming video diffusion model for real-time interactive world modeling with long-term geometric consistency, extending HunyuanWorld 1.0 beyond offline 3D world generation.",
      "highlights": [
        "WorldPlay-8B is based on HunyuanVideo",
        "A lighter WorldPlay-5B model and training code followed on January 6, 2026"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/HY-WorldPlay: HY-World 1.5: A Systematic Framework for Interactive World Modeling with Real-Time Latency and Geometric Consistency",
      "source_url": "https://github.com/Tencent-Hunyuan/HY-WorldPlay",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#hy-world-1-5-worldplay"
    },
    {
      "slug": "mistral-ocr-3",
      "date": "2025-12-17",
      "model": "Mistral OCR 3",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "OCR model that extracts text and embedded images from forms, scans, tables and handwriting, returning markdown with HTML-based table reconstruction.",
      "highlights": [
        "Powers the Document AI playground in Mistral AI Studio",
        "API model ID mistral-ocr-2512",
        "Major upgrade over Mistral OCR 2 on handwriting and low-quality scans"
      ],
      "source_title": "Introducing Mistral OCR 3 | Mistral AI",
      "source_url": "https://mistral.ai/news/mistral-ocr-3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#mistral-ocr-3"
    },
    {
      "slug": "flux-2-max",
      "date": "2025-12-16",
      "model": "FLUX.2 [max]",
      "lab": "Black Forest Labs",
      "lab_slug": "black-forest-labs",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Highest-quality FLUX.2 model, adding grounding search that generates images from real-time web information, alongside the strongest editing consistency and prompt following in the FLUX.2 family.",
      "highlights": [
        "Grounding search for current events, weather and news imagery",
        "Multi-reference editing with up to 10 input images",
        "Available in the BFL Playground and API"
      ],
      "source_title": "Release Notes - Black Forest Labs",
      "source_url": "https://docs.bfl.ai/release-notes",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#flux-2-max"
    },
    {
      "slug": "gpt-image-1-5",
      "date": "2025-12-16",
      "model": "GPT Image 1.5",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image",
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "GPT Image 1.5 is OpenAI's new flagship image generation model, powering a new version of ChatGPT Images and released in the API. OpenAI says it makes more precise edits that keep details intact and generates images up to 4x faster.",
      "highlights": [
        "API model ID gpt-image-1.5 (snapshot gpt-image-1.5-2025-12-16), plus chatgpt-image-latest",
        "Rolled out to all ChatGPT users as the model behind the new ChatGPT Images",
        "New Images creation space in ChatGPT; Business and Enterprise access came later"
      ],
      "source_title": "The new ChatGPT Images is here | OpenAI",
      "source_url": "https://openai.com/index/new-chatgpt-images-is-here/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#gpt-image-1-5"
    },
    {
      "slug": "longcat-video-avatar",
      "date": "2025-12-16",
      "model": "LongCat-Video-Avatar",
      "lab": "Meituan",
      "lab_slug": "meituan",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "audio"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "LongCat-Video-Avatar is a unified open model for expressive, highly dynamic audio-driven character video generation, built on Meituan's LongCat-Video foundation model and released with a technical report.",
      "highlights": [
        "Supports audio-text-to-video and audio-image-to-video generation",
        "Handles single- and multi-speaker audio inputs",
        "Weights released under MIT"
      ],
      "source_title": "GitHub - meituan-longcat/LongCat-Video",
      "source_url": "https://github.com/meituan-longcat/LongCat-Video",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#longcat-video-avatar"
    },
    {
      "slug": "mimo-v2-flash",
      "date": "2025-12-16",
      "model": "MiMo-V2-Flash",
      "lab": "Xiaomi",
      "lab_slug": "xiaomi",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "MoE: 309B total, 15B active",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "MiMo-V2-Flash is Xiaomi's open-weight mixture-of-experts model for reasoning, coding, and agentic tasks, using hybrid sliding-window and global attention plus multi-token prediction for fast, efficient inference.",
      "highlights": [
        "Hybrid attention at a 5:1 ratio with a 128-token sliding window",
        "Hybrid thinking mode that can be switched on or off",
        "Weights, inference code, and 3-layer MTP weights released under MIT",
        "API updates on January 12 and February 4, 2026 improved coding and tool calling"
      ],
      "source_title": "MiMo-V2-Flash | Xiaomi",
      "source_url": "https://mimo.xiaomi.com/mimo-v2-flash/index.html",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#mimo-v2-flash"
    },
    {
      "slug": "mistral-small-creative",
      "date": "2025-12-16",
      "model": "Mistral Small Creative",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Experimental Labs model from Mistral trained on curated data for creative writing, narrative generation, roleplay, character-driven dialogue and general-purpose conversational agents.",
      "highlights": [
        "API model ID labs-mistral-small-creative",
        "Mistral's model card lists December 11, 2025",
        "Deprecated March 31, 2026 in favor of Ministral 3 8B"
      ],
      "source_title": "Changelog | Mistral Docs",
      "source_url": "https://docs.mistral.ai/resources/changelogs",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#mistral-small-creative"
    },
    {
      "slug": "meta-sam-audio",
      "date": "2025-12-16",
      "model": "SAM Audio",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "audio",
        "text",
        "video"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": "500M to 3B",
      "model_license": "SAM License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "SAM Audio extends the Segment Anything approach to sound, a unified model that isolates specific sounds from audio mixtures using text, visual or time-span prompts.",
      "highlights": [
        "Prompts can be text, clicks on sounding objects in video, or marked time spans",
        "Released alongside the PE-AV perception encoder",
        "Audio prompts are not supported"
      ],
      "source_title": "Introducing SAM Audio: The First Unified Multimodal Model for Audio Separation",
      "source_url": "https://ai.meta.com/blog/sam-audio/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#meta-sam-audio"
    },
    {
      "slug": "seedance-1-5-pro",
      "date": "2025-12-16",
      "model": "Seedance 1.5 pro",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video",
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Seedance 1.5 pro is ByteDance Seed's audio-visual generation model, adding joint audio and video synthesis to Seedance with precise lip and audio synchronization, multilingual and dialect speech, and stronger narrative coherence.",
      "highlights": [
        "Generates synchronized audio and video together",
        "Supports multiple languages and Chinese dialects in generated speech",
        "Available in Jimeng AI and the Doubao app at launch",
        "Volcano Engine API version supports first and last frame control"
      ],
      "source_title": "Sound and Vision, All in One Take | The Official Release of Seedance 1.5 pro",
      "source_url": "https://seed.bytedance.com/en/blog/sound-and-vision-all-in-one-take-the-official-release-of-seedance-1-5-pro",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#seedance-1-5-pro"
    },
    {
      "slug": "bolmo",
      "date": "2025-12-15",
      "model": "Bolmo",
      "lab": "Ai2",
      "lab_slug": "ai2",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "1B and 7B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Family of byte-level language models created by byteifying Olmo 3 models, reusing their backbones while replacing subword tokenization with byte-level patching.",
      "highlights": [
        "Uses an mLSTM local encoder and a boundary predictor",
        "Built on the public Olmo 3 7B checkpoint",
        "Released under Apache 2.0"
      ],
      "source_title": "Introducing Bolmo: Byteifying the next generation of language models | Ai2",
      "source_url": "https://allenai.org/blog/bolmo",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#bolmo"
    },
    {
      "slug": "nemotron-3-nano",
      "date": "2025-12-15",
      "model": "Nemotron 3 Nano",
      "lab": "NVIDIA",
      "lab_slug": "nvidia",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": "MoE: 30B total, 3.5B active",
      "model_license": "NVIDIA Nemotron Open Model License",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "First model of the Nemotron 3 open family, a hybrid Mamba-Transformer mixture-of-experts model with a 1M-token context window for agentic workloads.",
      "highlights": [
        "Each MoE layer has 128 experts plus a shared expert",
        "Super and Ultra models were announced for the first half of 2026",
        "Available on Hugging Face and through inference providers"
      ],
      "source_title": "NVIDIA Debuts Nemotron 3 Family of Open Models | NVIDIA Newsroom",
      "source_url": "https://nvidianews.nvidia.com/news/nvidia-debuts-nemotron-3-family-of-open-models",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#nemotron-3-nano"
    },
    {
      "slug": "olmo-3-1",
      "date": "2025-12-12",
      "model": "Olmo 3.1",
      "lab": "Ai2",
      "lab_slug": "ai2",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "32B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Update to Olmo 3 with two new 32B checkpoints, Olmo 3.1 Think 32B and Olmo 3.1 Instruct 32B, plus refreshed RL Zero 7B models.",
      "highlights": [
        "Think 32B came from 21 more days of RL training on 224 GPUs",
        "Instruct 32B applies the Olmo 3 Instruct 7B recipe at larger scale",
        "Released under Apache 2.0"
      ],
      "source_title": "Olmo 3: Charting a path through the model flow to lead open-source AI | Ai2",
      "source_url": "https://allenai.org/blog/olmo3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#olmo-3-1"
    },
    {
      "slug": "glm-tts",
      "date": "2025-12-11",
      "model": "GLM-TTS",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "GLM-TTS is Z.ai's controllable, emotion-expressive zero-shot text-to-speech model, trained with multi-reward reinforcement learning. Z.ai launched it on its API alongside GLM-TTS-Clone for voice cloning from about 3 seconds of audio.",
      "highlights": [
        "Two-stage generation architecture",
        "Streaming and non-streaming interfaces",
        "Weights and inference scripts open-sourced under the MIT license"
      ],
      "source_title": "GLM-TTS (GitHub)",
      "source_url": "https://github.com/zai-org/GLM-TTS",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#glm-tts"
    },
    {
      "slug": "gpt-5-2",
      "date": "2025-12-11",
      "model": "GPT-5.2 (Instant, Thinking, and Pro)",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API",
        "Enterprise"
      ],
      "summary": "GPT-5.2 is the next model series in the GPT-5 line, released as Instant, Thinking, and Pro in ChatGPT and the API. OpenAI positions it for professional knowledge work such as spreadsheets, presentations, coding, and long-running agents.",
      "highlights": [
        "API IDs: gpt-5.2 (Thinking), gpt-5.2-chat-latest (Instant), and gpt-5.2-pro (Responses API)",
        "Adds a fifth reasoning effort level, xhigh, for GPT-5.2 Thinking and Pro",
        "Context: 400,000 tokens for gpt-5.2 and gpt-5.2-pro; 128,000 for gpt-5.2-chat-latest",
        "ChatGPT rollout began with paid plans, including Business and Enterprise"
      ],
      "source_title": "Introducing GPT-5.2 | OpenAI",
      "source_url": "https://openai.com/index/introducing-gpt-5-2/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#gpt-5-2"
    },
    {
      "slug": "molmo-2",
      "date": "2025-12-11",
      "model": "Molmo 2",
      "lab": "Ai2",
      "lab_slug": "ai2",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "4B, 7B and 8B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Successor to Ai2's Molmo that extends its grounded vision capabilities to video and multi-image understanding, adding video pointing and tracking, released with open weights.",
      "highlights": [
        "Molmo 2 8B and 4B are built on Qwen 3",
        "Molmo 2-O 7B is built on Olmo for a fully open stack",
        "Full training codebase released later"
      ],
      "source_title": "Molmo 2: State-of-the-art video understanding, pointing, and tracking | Ai2",
      "source_url": "https://allenai.org/blog/molmo2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#molmo-2"
    },
    {
      "slug": "cohere-rerank-4",
      "date": "2025-12-11",
      "model": "Rerank 4",
      "lab": "Cohere",
      "lab_slug": "cohere",
      "type": "reranker",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "score"
      ],
      "context_window": 32000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Fourth-generation reranker with a 32K-token context window, four times the previous generation, offered in Pro and Fast variants for enterprise AI search.",
      "highlights": [
        "Model IDs rerank-v4.0-pro and rerank-v4.0-fast",
        "Covers 100+ languages",
        "Key component of the North agent platform"
      ],
      "source_title": "Rerank 4: Cohere's Most Powerful Reranker Yet | Cohere",
      "source_url": "https://cohere.com/blog/rerank-4",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#cohere-rerank-4"
    },
    {
      "slug": "scail-preview",
      "date": "2025-12-11",
      "model": "SCAIL-Preview",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "SCAIL-Preview is the preview release of SCAIL, Z.ai's studio-grade character animation model that learns 3D-consistent pose representations in context. It animates a reference character from a driving pose sequence.",
      "highlights": [
        "Inference code released December 8, 2025; weights followed December 11",
        "Weights on Hugging Face and ModelScope",
        "Released under the MIT license"
      ],
      "source_title": "SCAIL: Towards Studio-Grade Character Animation (GitHub)",
      "source_url": "https://github.com/zai-org/SCAIL",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#scail-preview"
    },
    {
      "slug": "glm-asr-2512",
      "date": "2025-12-10",
      "model": "GLM-ASR-2512 and GLM-ASR-Nano-2512",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "1.5B (Nano)",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "GLM-ASR-2512 is Z.ai's speech recognition model on its API, released with GLM-ASR-Nano-2512, a 1.5B open-weights version. They support Mandarin, Cantonese, other dialects, and dozens of languages.",
      "highlights": [
        "Native custom dictionary support for domain terms",
        "GLM-ASR-Nano-2512 is aimed at challenging acoustic environments",
        "Nano weights on Hugging Face under the MIT license"
      ],
      "source_title": "Model and product release records (BigModel docs)",
      "source_url": "https://docs.bigmodel.cn/cn/update/new-releases",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#glm-asr-2512"
    },
    {
      "slug": "devstral-2",
      "date": "2025-12-09",
      "model": "Devstral 2",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "123B",
      "model_license": "Modified MIT License",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Next-generation 123B dense coding model for code agents, released under a modified MIT license together with the Mistral Vibe command-line agent.",
      "highlights": [
        "256K-token context window",
        "Deprecated May 22, 2026 in favor of Mistral Medium 3.5",
        "API model ID devstral-2512"
      ],
      "source_title": "Introducing: Devstral 2 and Mistral Vibe CLI. | Mistral AI",
      "source_url": "https://mistral.ai/news/devstral-2-vibe-cli",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#devstral-2"
    },
    {
      "slug": "devstral-small-2",
      "date": "2025-12-09",
      "model": "Devstral Small 2",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "24B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "24B coding model released alongside Devstral 2 under Apache 2.0, available through the API or deployable locally on consumer hardware.",
      "highlights": [
        "256K-token context window",
        "API model ID labs-devstral-small-2512",
        "Supports on-premises deployment and fine-tuning"
      ],
      "source_title": "Introducing: Devstral 2 and Mistral Vibe CLI. | Mistral AI",
      "source_url": "https://mistral.ai/news/devstral-2-vibe-cli",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#devstral-small-2"
    },
    {
      "slug": "nomos-1",
      "date": "2025-12-09",
      "model": "Nomos 1",
      "lab": "Nous Research",
      "lab_slug": "nous-research",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "30B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "30B open-weights mathematics model released by Nous Research in partnership with Hillclimb, described by Nous as a small but capable mathematician for problem solving.",
      "highlights": [
        "Focused on mathematical problem solving",
        "Built in partnership with Hillclimb",
        "Published on Hugging Face as nomos-1"
      ],
      "source_title": "Releases | Nous Research",
      "source_url": "https://nousresearch.com/releases",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#nomos-1"
    },
    {
      "slug": "autoglm-phone-9b",
      "date": "2025-12-08",
      "model": "AutoGLM-Phone-9B",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "9B",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "AutoGLM-Phone-9B is Z.ai's open phone agent model, released with the Open-AutoGLM framework for completing app tasks on a phone from natural language. Its architecture is identical to GLM-4.1V-9B-Thinking.",
      "highlights": [
        "Released with a multilingual variant, AutoGLM-Phone-9B-Multilingual",
        "Z.ai launched the AutoGLM-Phone framework on its API on December 11, 2025",
        "Weights under the MIT license"
      ],
      "source_title": "Open-AutoGLM (GitHub)",
      "source_url": "https://github.com/zai-org/Open-AutoGLM",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#autoglm-phone-9b"
    },
    {
      "slug": "glm-4-6v",
      "date": "2025-12-08",
      "model": "GLM-4.6V",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video",
        "file"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "106B (GLM-4.6V); 9B (GLM-4.6V-Flash)",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "GLM-4.6V is Z.ai's open-weights vision-language series, with GLM-4.6V for cloud use and GLM-4.6V-Flash for local, low-latency use. Z.ai says it integrates native function calling for the first time, passing images directly as tool parameters.",
      "highlights": [
        "Visual context window scaled to 128K tokens in training",
        "Generates interleaved image and text content from multimodal inputs",
        "Weights under the MIT license"
      ],
      "source_title": "GLM-4.6V: Open Source Multimodal Models with Native Tool Use",
      "source_url": "https://z.ai/blog/glm-4.6v",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#glm-4-6v"
    },
    {
      "slug": "longcat-image",
      "date": "2025-12-05",
      "model": "LongCat-Image",
      "lab": "Meituan",
      "lab_slug": "meituan",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "6B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "LongCat-Image is Meituan's open 6B-parameter image generation model, released together with a developer checkpoint and the LongCat-Image-Edit image editing model.",
      "highlights": [
        "6B parameters",
        "LongCat-Image, LongCat-Image-Dev, and LongCat-Image-Edit weights released together",
        "A distilled LongCat-Image-Edit-Turbo followed on February 3, 2026",
        "Licensed under Apache-2.0"
      ],
      "source_title": "GitHub - meituan-longcat/LongCat-Image",
      "source_url": "https://github.com/meituan-longcat/LongCat-Image",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#longcat-image"
    },
    {
      "slug": "gemini-3-deep-think",
      "date": "2025-12-04",
      "model": "Gemini 3 Deep Think",
      "lab": "Google",
      "lab_slug": "google",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Gemini 3 Deep Think is an enhanced reasoning mode built on Gemini 3 Pro for complex math, science and logic problems, rolled out to Google AI Ultra subscribers in the Gemini app.",
      "highlights": [
        "Introduced alongside Gemini 3 on Nov 18, 2025 as an enhanced reasoning mode coming soon to Ultra subscribers",
        "Received a major upgrade for science, research and engineering on Feb 12, 2026",
        "The February 2026 upgrade opened Gemini API access to select researchers and enterprises through an early access program"
      ],
      "source_title": "Gemini 3 Deep Think is now available",
      "source_url": "https://blog.google/products-and-platforms/products/gemini/gemini-3-deep-think/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#gemini-3-deep-think"
    },
    {
      "slug": "hermes-4-3",
      "date": "2025-12-03",
      "model": "Hermes 4.3",
      "lab": "Nous Research",
      "lab_slug": "nous-research",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 512000,
      "parameters": "36B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Update to the Hermes series that nearly matches Hermes 4 70B at half the size, post-trained entirely on Nous Research's decentralized Psyche network.",
      "highlights": [
        "Trained with extended context up to 512K tokens",
        "Released as Hermes-4.3-Seed-36B",
        "Psyche-trained version outperformed a centrally trained version on downstream tasks"
      ],
      "source_title": "Introducing Hermes 4.3: Local Intelligence Globally Trained | Nous Research",
      "source_url": "https://nousresearch.com/introducing-hermes-4-3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#hermes-4-3"
    },
    {
      "slug": "kling-video-2-6",
      "date": "2025-12-03",
      "model": "Kling Video 2.6",
      "lab": "Kuaishou",
      "lab_slug": "kuaishou",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video",
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Kling Video 2.6 is Kling's first model with simultaneous audio-visual generation, producing visuals together with voiceovers, sound effects, and ambient sound in a single pass from text or image prompts.",
      "highlights": [
        "Generates speech, sound effects, and ambient audio with the video",
        "Supports Chinese and English voice generation",
        "Generates clips up to 10 seconds long"
      ],
      "source_title": "Kling AI Launches Video 2.6 Model with \"Simultaneous Audio-Visual Generation\" Capability, Redefining AI Video Creation Workflow",
      "source_url": "https://ir.kuaishou.com/news-releases/news-release-details/kling-ai-launches-video-26-model-simultaneous-audio-visual",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#kling-video-2-6"
    },
    {
      "slug": "seedream-4-5",
      "date": "2025-12-03",
      "model": "Seedream 4.5",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Seedream 4.5 is an upgrade to Seedream 4.0 that scales up the model to improve subject consistency in multi-image editing, preserve reference image details, and strengthen typography and dense text rendering.",
      "highlights": [
        "Supports up to 10 reference images for multi-image fusion",
        "Stylized output up to 4K resolution",
        "Launched in open beta on BytePlus ModelArk"
      ],
      "source_title": "BytePlus Unveils Seedream 4.5: Precision-Focused Upgrade Delivering Sharper Visuals, Smarter Control, and 4K Creative Consistency",
      "source_url": "https://www.byteplus.com/en/blog/seedream4-5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#seedream-4-5"
    },
    {
      "slug": "amazon-nova-2-lite",
      "date": "2025-12-02",
      "model": "Amazon Nova 2 Lite",
      "lab": "Amazon",
      "lab_slug": "amazon",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Amazon Nova 2 Lite is a fast, cost-effective reasoning model for everyday workloads, with optional extended thinking at three intensity levels, multimodal input and a one-million-token context window.",
      "highlights": [
        "Extended thinking is off by default, with low, medium and high levels",
        "One-million-token context window and up to 64K output tokens",
        "Built-in web grounding and code interpreter tools"
      ],
      "source_title": "Introducing Amazon Nova 2 Lite, a fast, cost-effective reasoning model",
      "source_url": "https://aws.amazon.com/blogs/aws/introducing-amazon-nova-2-lite-a-fast-cost-effective-reasoning-model/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#amazon-nova-2-lite"
    },
    {
      "slug": "amazon-nova-2-sonic",
      "date": "2025-12-02",
      "model": "Amazon Nova 2 Sonic",
      "lab": "Amazon",
      "lab_slug": "amazon",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": 1000000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Amazon Nova 2 Sonic is Amazon's second-generation speech-to-speech model for conversational AI, adding Portuguese and Hindi, polyglot voices, configurable turn-taking, crossmodal text and voice input, and asynchronous tool calling.",
      "highlights": [
        "Supports seven languages: English, French, Italian, German, Spanish, Portuguese and Hindi",
        "One-million-token context window per the Bedrock model card",
        "Integrates with Amazon Connect and telephony providers"
      ],
      "source_title": "Introducing Amazon Nova 2 Sonic: Our new speech-to-speech model for conversational AI",
      "source_url": "https://aws.amazon.com/blogs/aws/introducing-amazon-nova-2-sonic-next-generation-speech-to-speech-model-for-conversational-ai/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#amazon-nova-2-sonic"
    },
    {
      "slug": "browsesafe",
      "date": "2025-12-02",
      "model": "BrowseSafe",
      "lab": "Perplexity",
      "lab_slug": "perplexity",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 16384,
      "parameters": null,
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "An open prompt-injection detection model fine-tuned from Qwen3-30B-A3B that reads a web page's HTML and answers whether it contains malicious instructions aimed at a browser agent, released with the BrowseSafe-Bench benchmark.",
      "highlights": [
        "Outputs a single yes or no token for each page",
        "Context length up to 16,384 tokens per the model card",
        "BrowseSafe-Bench spans 11 attack types and 9 injection strategies",
        "Built for Perplexity's Comet browser; weights on Hugging Face under the MIT license"
      ],
      "source_title": "Building Safer AI Browsers with BrowseSafe",
      "source_url": "https://www.perplexity.ai/hub/blog/building-safer-ai-browsers-with-browsesafe",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#browsesafe"
    },
    {
      "slug": "ministral-3",
      "date": "2025-12-02",
      "model": "Ministral 3 (3B, 8B, 14B)",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "3B, 8B and 14B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Family of three dense small models released with Mistral Large 3, each with image understanding and offered in base, instruct and reasoning variants.",
      "highlights": [
        "256K-token context window",
        "Optimized for edge hardware including DGX Spark, RTX PCs and Jetson",
        "API model IDs ministral-3b-2512, ministral-8b-2512 and ministral-14b-2512"
      ],
      "source_title": "Introducing Mistral 3 | Mistral AI",
      "source_url": "https://mistral.ai/news/mistral-3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#ministral-3"
    },
    {
      "slug": "mistral-large-3",
      "date": "2025-12-02",
      "model": "Mistral Large 3",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "MoE: 675B total, 41B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Mistral's first mixture-of-experts model since Mixtral, a sparse open-weight multimodal model trained from scratch on 3,000 NVIDIA H200 GPUs and released under Apache 2.0.",
      "highlights": [
        "Base and instruct versions released",
        "NVFP4 checkpoint for Blackwell NVL72 and single 8-GPU nodes",
        "API model ID mistral-large-2512"
      ],
      "source_title": "Introducing Mistral 3 | Mistral AI",
      "source_url": "https://mistral.ai/news/mistral-3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#mistral-large-3"
    },
    {
      "slug": "deepseek-v3-2",
      "date": "2025-12-01",
      "model": "DeepSeek-V3.2",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "DeepSeek-V3.2 is the official successor to V3.2-Exp, a reasoning-first model built for agents. It is DeepSeek's first model to integrate thinking directly into tool use, supporting tool calls in both thinking and non-thinking modes.",
      "highlights": [
        "Built on DeepSeek Sparse Attention for efficient long-context inference",
        "Trained with agent data synthesized across more than 1,800 environments and 85,000 complex instructions",
        "Powers both deepseek-chat (non-thinking) and deepseek-reasoner (thinking) on the API"
      ],
      "source_title": "DeepSeek-V3.2 Release",
      "source_url": "https://api-docs.deepseek.com/news/news251201",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#deepseek-v3-2"
    },
    {
      "slug": "deepseek-v3-2-speciale",
      "date": "2025-12-01",
      "model": "DeepSeek-V3.2-Speciale",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "DeepSeek-V3.2-Speciale is a high-compute reasoning variant of DeepSeek-V3.2 that pushes maximum reasoning capability at the cost of higher token usage. DeepSeek reports gold-level results at IMO, CMO, the ICPC World Finals, and IOI 2025.",
      "highlights": [
        "Served through a temporary API endpoint available until December 15, 2025",
        "No tool calling support at launch, offered for community evaluation and research",
        "Weights on Hugging Face under the MIT License"
      ],
      "source_title": "DeepSeek-V3.2 Release",
      "source_url": "https://api-docs.deepseek.com/news/news251201",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#deepseek-v3-2-speciale"
    },
    {
      "slug": "runway-gen-4-5",
      "date": "2025-12-01",
      "model": "Gen-4.5",
      "lab": "Runway",
      "lab_slug": "runway",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Video generation model advancing pre-training data efficiency and post-training, with improved motion, physical accuracy, prompt adherence and temporal consistency at Gen-4's speed and efficiency.",
      "highlights": [
        "Available for all paid plans from December 11, 2025",
        "Image to video added January 21, 2026",
        "Trained and served on NVIDIA Hopper and Blackwell GPUs"
      ],
      "source_title": "Runway Gen-4.5: State-of-the-Art AI Video Generation",
      "source_url": "https://runway.com/research/introducing-runway-gen-4.5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#runway-gen-4-5"
    },
    {
      "slug": "kling-o1",
      "date": "2025-12-01",
      "model": "Kling O1",
      "lab": "Kuaishou",
      "lab_slug": "kuaishou",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "video",
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Kling O1 is a unified multimodal video model that combines reference-based generation, text-to-video, start and end frames, in-painting, restyling, and shot extension in one engine built on a Multimodal Visual Language framework.",
      "highlights": [
        "Accepts text, image, video, and subject references as prompts",
        "Supports conversational video editing without manual masking",
        "Kling Image O1 image model followed on December 2, 2025"
      ],
      "source_title": "Kling O1 Launches as the World's First Unified Multimodal Video Model",
      "source_url": "https://ir.kuaishou.com/news-releases/news-release-details/kling-o1-launches-worlds-first-unified-multimodal-video-model-0",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/12#kling-o1"
    },
    {
      "slug": "deepseekmath-v2",
      "date": "2025-11-27",
      "model": "DeepSeekMath-V2",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "DeepSeekMath-V2 is an open-weights model for self-verifiable mathematical reasoning and theorem proving, built on DeepSeek-V3.2-Exp-Base. DeepSeek reports gold-level scores on IMO 2025 and CMO 2024 with scaled test-time compute.",
      "highlights": [
        "Targets rigorous step-by-step proofs rather than final-answer accuracy alone",
        "Built on top of DeepSeek-V3.2-Exp-Base",
        "Weights released under the Apache 2.0 license"
      ],
      "source_title": "deepseek-ai/DeepSeek-Math-V2 (Hugging Face)",
      "source_url": "https://huggingface.co/deepseek-ai/DeepSeek-Math-V2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/11#deepseekmath-v2"
    },
    {
      "slug": "step1x-edit-v1-2",
      "date": "2025-11-26",
      "model": "Step1X-Edit v1.2",
      "lab": "StepFun",
      "lab_slug": "stepfun",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Step1X-Edit v1.2, described in its paper as ReasonEdit-S, is a version of StepFun's open image editing model with native reasoning-based editing, following a v1.2 preview released in September 2025.",
      "highlights": [
        "Adds native reasoning edit ability to the Step1X-Edit line",
        "A step1x-edit-v1p2-preview checkpoint was released on September 8, 2025",
        "Weights on Hugging Face under Apache-2.0"
      ],
      "source_title": "GitHub - stepfun-ai/Step1X-Edit",
      "source_url": "https://github.com/stepfun-ai/Step1X-Edit",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/11#step1x-edit-v1-2"
    },
    {
      "slug": "flux-2-dev",
      "date": "2025-11-25",
      "model": "FLUX.2 [dev]",
      "lab": "Black Forest Labs",
      "lab_slug": "black-forest-labs",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "32B",
      "model_license": "FLUX Non-Commercial License",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "32B-parameter open-weight model derived from the FLUX.2 base model, combining text-to-image synthesis and multi-reference image editing in a single checkpoint.",
      "highlights": [
        "Optimized FP8 reference implementation for GeForce RTX GPUs with NVIDIA and ComfyUI",
        "Released alongside the FLUX.2 VAE under Apache 2.0",
        "Hosted by partners including fal, Replicate, Together AI and Cloudflare"
      ],
      "source_title": "FLUX.2: Frontier Visual Intelligence",
      "source_url": "https://bfl.ai/blog/flux-2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/11#flux-2-dev"
    },
    {
      "slug": "flux-2-pro-flex",
      "date": "2025-11-25",
      "model": "FLUX.2 [pro] and [flex]",
      "lab": "Black Forest Labs",
      "lab_slug": "black-forest-labs",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Cloud platforms"
      ],
      "summary": "Second-generation FLUX image models that combine generation and multi-reference editing at up to 4MP; [pro] targets production speed and quality while [flex] exposes steps and guidance for fine control and typography.",
      "highlights": [
        "Reference up to 10 images for character, product and style consistency",
        "Hex-code color control and JSON-style structured prompts",
        "Available in the BFL Playground, BFL API and launch partners"
      ],
      "source_title": "FLUX.2: Frontier Visual Intelligence",
      "source_url": "https://bfl.ai/blog/flux-2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/11#flux-2-pro-flex"
    },
    {
      "slug": "hunyuanocr",
      "date": "2025-11-25",
      "model": "HunyuanOCR",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "1B",
      "model_license": "Tencent Hunyuan Community License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "HunyuanOCR is an open end-to-end OCR vision-language model built on Hunyuan's native multimodal architecture, handling document parsing, text spotting, information extraction, and text-image translation in one lightweight model.",
      "highlights": [
        "1B parameters",
        "Supports over 100 languages, including mixed-language documents"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/HunyuanOCR: HunyuanOCR-1.5: Making Lightweight OCR VLMs Faster and Better",
      "source_url": "https://github.com/Tencent-Hunyuan/HunyuanOCR",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/11#hunyuanocr"
    },
    {
      "slug": "claude-opus-4-5",
      "date": "2025-11-24",
      "model": "Claude Opus 4.5",
      "lab": "Anthropic",
      "lab_slug": "anthropic",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 200000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Cloud platforms"
      ],
      "summary": "Claude Opus 4.5 is Anthropic's Opus model for coding, agents, and computer use, with improvements on everyday tasks such as deep research, slides, and spreadsheets. It launched with an effort parameter in beta on the API.",
      "highlights": [
        "API model ID claude-opus-4-5-20251101",
        "200K-token context window and 64K max output tokens",
        "Available in the Claude apps, on the Claude API, and on all three major cloud platforms"
      ],
      "source_title": "Introducing Claude Opus 4.5",
      "source_url": "https://www.anthropic.com/news/claude-opus-4-5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/11#claude-opus-4-5"
    },
    {
      "slug": "fara-7b",
      "date": "2025-11-24",
      "model": "Fara-7B",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "7B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Fara-7B is Microsoft's first agentic small language model for computer use, taking screenshots and text context as input and directly predicting thoughts and grounded actions to complete web tasks.",
      "highlights": [
        "7B parameters with a 128K context length",
        "Builds on Qwen 2.5 VL",
        "Available on Hugging Face and Microsoft Foundry"
      ],
      "source_title": "microsoft/Fara-7B on Hugging Face",
      "source_url": "https://huggingface.co/microsoft/Fara-7B",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/11#fara-7b"
    },
    {
      "slug": "hunyuanvideo-1-5",
      "date": "2025-11-20",
      "model": "HunyuanVideo 1.5",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": "8.3B",
      "model_license": "Tencent Hunyuan Community License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "HunyuanVideo 1.5 is a lightweight open video generation model with 8.3B parameters for text-to-video and image-to-video generation, designed to run on consumer-grade GPUs.",
      "highlights": [
        "8.3B-parameter Diffusion Transformer with a 3D causal VAE",
        "Few-step super-resolution network upscales output to 1080p",
        "Step-distilled 480p image-to-video model and training code added December 5, 2025"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/HunyuanVideo-1.5: HunyuanVideo-1.5: A leading lightweight video generation model",
      "source_url": "https://github.com/Tencent-Hunyuan/HunyuanVideo-1.5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/11#hunyuanvideo-1-5"
    },
    {
      "slug": "keye-vl-671b-a37b",
      "date": "2025-11-20",
      "model": "Keye-VL-671B-A37B",
      "lab": "Kuaishou",
      "lab_slug": "kuaishou",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "MoE: 671B total, 37B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Keye-VL-671B-A37B is the largest model in Kuaishou's Keye multimodal series, combining the Keye-VL-1.5 vision encoder with upgraded pre-training and post-training for text, image, and video understanding.",
      "highlights": [
        "Reuses the Keye-VL-1.5 vision encoder",
        "Synthetic chain-of-thought data used during continual pre-training",
        "Weights released under Apache-2.0"
      ],
      "source_title": "GitHub - Kwai-Keye/Keye",
      "source_url": "https://github.com/Kwai-Keye/Keye",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/11#keye-vl-671b-a37b"
    },
    {
      "slug": "gemini-3-pro-image",
      "date": "2025-11-20",
      "model": "Nano Banana Pro (Gemini 3 Pro Image)",
      "lab": "Google",
      "lab_slug": "google",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text",
        "image"
      ],
      "context_window": 65536,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms",
        "Consumer app",
        "Enterprise"
      ],
      "summary": "Nano Banana Pro is Google DeepMind's image generation and editing model built on Gemini 3 Pro, adding studio-quality creative controls, legible text in multiple languages and grounding in real-time information.",
      "highlights": [
        "Up to 4K resolution output with localized editing, camera angle, focus and color grading controls",
        "Rolled out in the Gemini app, Google Ads, Slides, Vids, Flow, the Gemini API and Vertex AI",
        "Stable gemini-3-pro-image became generally available on May 28, 2026"
      ],
      "source_title": "Nano Banana Pro: Gemini 3 Pro Image model from Google DeepMind",
      "source_url": "https://blog.google/innovation-and-ai/products/nano-banana-pro/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/11#gemini-3-pro-image"
    },
    {
      "slug": "olmo-3",
      "date": "2025-11-20",
      "model": "Olmo 3",
      "lab": "Ai2",
      "lab_slug": "ai2",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "7B and 32B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Third generation of fully open Olmo models, released as Base, Think, Instruct and RL Zero variants with checkpoints from every major stage of training.",
      "highlights": [
        "Pretrained on the new Dolma 3 data",
        "Releases base, mid-trained, long-context and post-training checkpoints",
        "Released under Apache 2.0"
      ],
      "source_title": "Olmo 3: Charting a path through the model flow to lead open-source AI | Ai2",
      "source_url": "https://allenai.org/blog/olmo3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/11#olmo-3"
    },
    {
      "slug": "gpt-5-1-codex-max",
      "date": "2025-11-19",
      "model": "GPT-5.1-Codex-Max",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "coding",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 400000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "Enterprise"
      ],
      "summary": "GPT-5.1-Codex-Max is an agentic coding model for long-running, project-scale work in Codex. It is OpenAI's first model natively trained to operate across multiple context windows through compaction, and it replaced GPT-5.1-Codex as the Codex default.",
      "highlights": [
        "Available in Codex with ChatGPT Plus, Pro, Business, Edu, and Enterprise plans",
        "API model ID gpt-5.1-codex-max reached the Responses API on December 4, 2025",
        "First model OpenAI trained to operate in Windows environments",
        "400,000-token context window per the API model page"
      ],
      "source_title": "Building more with GPT-5.1-Codex-Max | OpenAI",
      "source_url": "https://openai.com/index/gpt-5-1-codex-max/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/11#gpt-5-1-codex-max"
    },
    {
      "slug": "grok-4-1-fast",
      "date": "2025-11-19",
      "model": "Grok 4.1 Fast",
      "lab": "xAI",
      "lab_slug": "xai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 2000000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Grok 4.1 Fast is xAI's tool-calling model for agentic API use cases such as customer support and finance, trained with long-horizon reinforcement learning to hold performance across its 2M-token context window.",
      "highlights": [
        "API model IDs: grok-4-1-fast-reasoning and grok-4-1-fast-non-reasoning",
        "Launched with the Agent Tools API (X data, web search, remote code execution)",
        "Retired from the API on May 15, 2026, with requests redirected to grok-4.3"
      ],
      "source_title": "Grok 4.1 Fast and Agent Tools API | SpaceXSI",
      "source_url": "https://x.ai/news/grok-4-1-fast",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/11#grok-4-1-fast"
    },
    {
      "slug": "mimo-embodied",
      "date": "2025-11-19",
      "model": "MiMo-Embodied",
      "lab": "Xiaomi",
      "lab_slug": "xiaomi",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "7B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "MiMo-Embodied is an open-source cross-embodied vision-language model that Xiaomi describes as the first to integrate autonomous driving with embodied AI tasks such as task planning, affordance prediction, and spatial understanding.",
      "highlights": [
        "Released as MiMo-Embodied-7B on Hugging Face",
        "Evaluated on 17 embodied AI and 12 autonomous driving benchmarks",
        "Code and evaluation configs on GitHub"
      ],
      "source_title": "XiaomiMiMo/MiMo-Embodied-7B",
      "source_url": "https://huggingface.co/XiaomiMiMo/MiMo-Embodied-7B",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/11#mimo-embodied"
    },
    {
      "slug": "meta-sam-3",
      "date": "2025-11-19",
      "model": "SAM 3",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "image",
        "video"
      ],
      "context_window": null,
      "parameters": "848M",
      "model_license": "SAM License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "SAM 3 is a unified model for detecting, segmenting and tracking objects in images and video, adding promptable concept segmentation from short text phrases or image exemplars.",
      "highlights": [
        "Finds all instances of a concept from a noun-phrase or exemplar prompt",
        "Weights, the SA-Co benchmark and a paper were released",
        "Available to try in the Segment Anything Playground"
      ],
      "source_title": "New Segment Anything Models Make it Easier to Detect Objects and Create 3D Reconstructions",
      "source_url": "https://about.fb.com/news/2025/11/new-sam-models-detect-objects-create-3d-reconstructions/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/11#meta-sam-3"
    },
    {
      "slug": "meta-sam-3d",
      "date": "2025-11-19",
      "model": "SAM 3D (Objects and Body)",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "image"
      ],
      "output_modalities": [
        "3d"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "SAM License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "SAM 3D is a pair of models that reconstruct 3D objects, scenes and human bodies from a single image: SAM 3D Objects for objects and scenes and SAM 3D Body for human pose and shape.",
      "highlights": [
        "Checkpoints and inference code released on GitHub",
        "MHR, the parametric human model used by SAM 3D Body, is released under a permissive commercial license",
        "Available to try in the Segment Anything Playground"
      ],
      "source_title": "Introducing SAM 3D: Powerful 3D Reconstruction for Physical World Images",
      "source_url": "https://ai.meta.com/blog/sam-3d/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/11#meta-sam-3d"
    },
    {
      "slug": "step-audio-r1",
      "date": "2025-11-19",
      "model": "Step-Audio-R1",
      "lab": "StepFun",
      "lab_slug": "stepfun",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "audio",
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Step-Audio-R1 is StepFun's audio reasoning model, applying chain-of-thought reasoning to audio understanding tasks. StepFun published the technical report and demo on November 19, 2025, and released the weights on November 27.",
      "highlights": [
        "Technical report and demo page released November 19, 2025",
        "Inference code and weights released November 27, 2025 on Hugging Face and ModelScope",
        "Repository code released under Apache-2.0"
      ],
      "source_title": "GitHub - stepfun-ai/Step-Audio-R1",
      "source_url": "https://github.com/stepfun-ai/Step-Audio-R1",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/11#step-audio-r1"
    },
    {
      "slug": "gemini-3-pro",
      "date": "2025-11-18",
      "model": "Gemini 3 Pro",
      "lab": "Google",
      "lab_slug": "google",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1048576,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms",
        "Consumer app"
      ],
      "summary": "Gemini 3 Pro was the first Gemini 3 model, a reasoning and multimodal understanding model with strong agentic and coding capabilities, released in preview across Google's consumer, developer and enterprise products.",
      "highlights": [
        "Shipped on day one in AI Mode in Google Search, a first for a Gemini model",
        "Preview model gemini-3-pro-preview introduced thinking levels, thought signatures and media resolution controls",
        "Input token limit of 1,048,576 per the Gemini API model page",
        "Replaced by Gemini 3.1 Pro Preview, with gemini-3-pro-preview shut down on Mar 9, 2026"
      ],
      "source_title": "Gemini 3: Introducing the latest Gemini AI model from Google",
      "source_url": "https://blog.google/products-and-platforms/products/gemini/gemini-3/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/11#gemini-3-pro"
    },
    {
      "slug": "grok-4-1",
      "date": "2025-11-17",
      "model": "Grok 4.1",
      "lab": "xAI",
      "lab_slug": "xai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Grok 4.1 is an update focused on real-world usability, with better creative, emotional, and collaborative interaction, trained on Grok 4's reinforcement learning infrastructure using agentic reasoning models as reward models.",
      "highlights": [
        "Available to all users on grok.com, X, and the iOS and Android apps",
        "Rolled out in Auto mode and selectable as Grok 4.1 in the model picker",
        "Thinking (code name quasarflux) and non-reasoning (code name tensor) modes",
        "Preceded by a two-week silent rollout on production traffic"
      ],
      "source_title": "Grok 4.1 | SpaceXSI",
      "source_url": "https://x.ai/news/grok-4-1",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/11#grok-4-1"
    },
    {
      "slug": "depth-anything-3",
      "date": "2025-11-14",
      "model": "Depth Anything 3",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "image",
        "video"
      ],
      "output_modalities": [
        "3d"
      ],
      "context_window": null,
      "parameters": "0.08B to 1.40B (model series)",
      "model_license": "Apache-2.0 (Small, Base, Metric, Mono); CC-BY-NC-4.0 (Large, Giant, Nested)",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Depth Anything 3 is an open visual geometry model family that uses a single plain transformer to predict depth and camera pose from any number of views, extending monocular depth estimation to multi-view spatial reconstruction.",
      "highlights": [
        "Main series in Giant, Large, Base, and Small sizes",
        "Also released metric-depth, monocular, and nested variants",
        "Paper, code, and models released together on November 14, 2025"
      ],
      "source_title": "GitHub - ByteDance-Seed/Depth-Anything-3: Depth Anything 3",
      "source_url": "https://github.com/ByteDance-Seed/Depth-Anything-3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/11#depth-anything-3"
    },
    {
      "slug": "ernie-5-0",
      "date": "2025-11-13",
      "model": "ERNIE 5.0",
      "lab": "Baidu",
      "lab_slug": "baidu",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "context_window": null,
      "parameters": "MoE: 2.4T total, about 3% active per token",
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "ERNIE 5.0 is Baidu's natively omni-modal foundation model that jointly models text, images, audio, and video in a unified autoregressive framework for both understanding and generation. Baidu unveiled a preview at Baidu World 2025.",
      "highlights": [
        "Preview offered through ERNIE Bot and Baidu AI Cloud's Qianfan platform at announcement",
        "Official version launched January 22, 2026, with 2.4 trillion parameters",
        "Mixture-of-experts design that activates about 3% of parameters per token"
      ],
      "source_title": "Baidu Unveils ERNIE 5.0 and a Series of AI Applications at Baidu World 2025, Ramps Up Global Push",
      "source_url": "https://www.prnewswire.com/news-releases/baidu-unveils-ernie-5-0-and-a-series-of-ai-applications-at-baidu-world-2025--ramps-up-global-push-302614531.html",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/11#ernie-5-0"
    },
    {
      "slug": "gpt-5-1-codex",
      "date": "2025-11-13",
      "model": "GPT-5.1-Codex and GPT-5.1-Codex-Mini",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "coding",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 400000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "GPT-5.1-Codex and GPT-5.1-Codex-Mini are versions of GPT-5.1 optimized for long-running, agentic coding tasks in Codex or similar coding agents, released in the API and in Codex alongside the GPT-5.1 developer launch.",
      "highlights": [
        "API model IDs gpt-5.1-codex and gpt-5.1-codex-mini",
        "Codex CLI and IDE extension defaulted to gpt-5.1-codex on macOS and Linux",
        "400,000-token context window for both, per the API model pages"
      ],
      "source_title": "Introducing GPT-5.1 for developers | OpenAI",
      "source_url": "https://openai.com/index/gpt-5-1-for-developers/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/11#gpt-5-1-codex"
    },
    {
      "slug": "gpt-5-1",
      "date": "2025-11-12",
      "model": "GPT-5.1 (Instant and Thinking)",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API",
        "Enterprise"
      ],
      "summary": "GPT-5.1 upgrades the GPT-5 series with GPT-5.1 Instant, which OpenAI describes as warmer and better at following instructions, and GPT-5.1 Thinking, which adapts its thinking time to task complexity. ChatGPT also gained new tone controls.",
      "highlights": [
        "API IDs gpt-5.1 (Thinking) and gpt-5.1-chat-latest (Instant), released in the API on November 13, 2025",
        "API adds a 'none' reasoning setting for low-latency responses",
        "Context: 400,000 tokens for gpt-5.1; 128,000 for gpt-5.1-chat-latest",
        "GPT-5.1 Pro replaced GPT-5 Pro in ChatGPT on November 19, 2025"
      ],
      "source_title": "GPT-5.1: A smarter, more conversational ChatGPT | OpenAI",
      "source_url": "https://openai.com/index/gpt-5-1/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/11#gpt-5-1"
    },
    {
      "slug": "ernie-4-5-vl-28b-a3b-thinking",
      "date": "2025-11-11",
      "model": "ERNIE-4.5-VL-28B-A3B-Thinking",
      "lab": "Baidu",
      "lab_slug": "baidu",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "MoE: 28B total, 3B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "ERNIE-4.5-VL-28B-A3B-Thinking is an open-weights multimodal reasoning upgrade of ERNIE-4.5-VL-28B-A3B, adding extended mid-training, reinforcement learning, and thinking with images that zooms into regions and calls image search.",
      "highlights": [
        "Activates 3B parameters per token",
        "Strengthened visual grounding plus video temporal awareness and event localization",
        "Weights on Hugging Face; runs with transformers, vLLM, and FastDeploy"
      ],
      "source_title": "ERNIE-4.5-VL-28B-A3B-Thinking: A Breakthrough in Multimodal AI",
      "source_url": "https://ernie.baidu.com/research/posts/ernie-4.5-vl-28b-a3b-thinking/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/11#ernie-4-5-vl-28b-a3b-thinking"
    },
    {
      "slug": "elevenlabs-scribe-v2-realtime",
      "date": "2025-11-11",
      "model": "Scribe v2 Realtime",
      "lab": "ElevenLabs",
      "lab_slug": "elevenlabs",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Low-latency speech-to-text model for voice agents, meeting assistants and live captioning, transcribing in under 150ms across 90 languages with next-word and punctuation prediction.",
      "highlights": [
        "Under 150ms latency",
        "Automatic language detection with mid-conversation switching",
        "Integrated into ElevenLabs Agents"
      ],
      "source_title": "Introducing Scribe v2 Realtime",
      "source_url": "https://elevenlabs.io/blog/introducing-scribe-v2-realtime",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/11#elevenlabs-scribe-v2-realtime"
    },
    {
      "slug": "meta-omnilingual-asr",
      "date": "2025-11-10",
      "model": "Omnilingual ASR",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "300M to 7B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Omnilingual ASR is a suite of open speech recognition models from Meta FAIR covering more than 1,600 languages, including 500 low-resource languages not previously transcribed by AI.",
      "highlights": [
        "Models range from 300M versions for low-power devices to 7B models",
        "Can extend to new languages from a few in-context audio-text examples",
        "Released with the Omnilingual ASR Corpus covering 350 underserved languages"
      ],
      "source_title": "Omnilingual ASR: Advancing Automatic Speech Recognition for 1,600+ Languages",
      "source_url": "https://ai.meta.com/blog/omnilingual-asr-advancing-automatic-speech-recognition/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/11#meta-omnilingual-asr"
    },
    {
      "slug": "gpt-5-codex-mini",
      "date": "2025-11-07",
      "model": "GPT-5-Codex-Mini",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "coding",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "GPT-5-Codex-Mini is a smaller, more cost-effective but less capable version of GPT-5-Codex, added as a model option in the Codex CLI and IDE extension to stretch ChatGPT subscription usage.",
      "highlights": [
        "Model ID gpt-5-codex-mini, selectable in Codex CLI and the IDE extension",
        "Provides roughly 4x more usage than GPT-5-Codex within a ChatGPT subscription",
        "Codex suggests switching to it at 90% of the 5-hour usage limit"
      ],
      "source_title": "ChatGPT & Codex changelog | ChatGPT Learn",
      "source_url": "https://learn.chatgpt.com/docs/changelog",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/11#gpt-5-codex-mini"
    },
    {
      "slug": "kimi-k2-thinking",
      "date": "2025-11-06",
      "model": "Kimi K2 Thinking",
      "lab": "Moonshot AI",
      "lab_slug": "moonshot-ai",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "MoE: 1T total, 32B active",
      "model_license": "Modified MIT License",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "Kimi K2 Thinking is Moonshot AI's open-weights thinking agent model, which reasons step by step while using tools. Moonshot says it can execute 200 to 300 sequential tool calls without human intervention.",
      "highlights": [
        "Scales both thinking tokens and tool-calling steps for test-time scaling",
        "Live in chat mode on kimi.com and through the Kimi K2 Thinking API",
        "Released with a Turbo variant on the Kimi Open Platform",
        "Weights and code under the Modified MIT License"
      ],
      "source_title": "Introducing Kimi K2 Thinking",
      "source_url": "https://www.kimi.com/en/blog/kimi-k2-thinking",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/11#kimi-k2-thinking"
    },
    {
      "slug": "step-audio-editx",
      "date": "2025-11-06",
      "model": "Step-Audio-EditX",
      "lab": "StepFun",
      "lab_slug": "stepfun",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "audio",
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": "3B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Step-Audio-EditX is a 3B-parameter LLM-based audio editing model trained with reinforcement learning to edit emotion, speaking style, and paralinguistic features of speech, alongside zero-shot text-to-speech.",
      "highlights": [
        "Technical report released November 6, 2025; weights followed on November 12",
        "Later updates added polyphonic pronunciation control plus Japanese and Korean support",
        "Repository released under Apache-2.0"
      ],
      "source_title": "GitHub - stepfun-ai/Step-Audio-EditX",
      "source_url": "https://github.com/stepfun-ai/Step-Audio-EditX",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/11#step-audio-editx"
    },
    {
      "slug": "minimax-music-2-0",
      "date": "2025-10-31",
      "model": "MiniMax Music 2.0",
      "lab": "MiniMax",
      "lab_slug": "minimax",
      "type": "music",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "MiniMax Music 2.0 is a music generation model with more lifelike vocals across singing styles and finer instrumental control, generating structurally complete songs with verses, choruses, and bridges of up to five minutes.",
      "highlights": [
        "Supports male-female duets and a cappella",
        "Covers genres including pop, jazz, blues, rock, and folk"
      ],
      "source_title": "Making Music Creation Accessible to Everyone",
      "source_url": "https://www.minimax.io/news/minimax-music-20",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/10#minimax-music-2-0"
    },
    {
      "slug": "kimi-linear",
      "date": "2025-10-30",
      "model": "Kimi Linear",
      "lab": "Moonshot AI",
      "lab_slug": "moonshot-ai",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": "MoE: 48B total, 3B active",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Kimi Linear is a hybrid linear attention architecture built on Kimi Delta Attention, a refined version of Gated DeltaNet. Moonshot released 48B base and instruct checkpoints trained on 5.7T tokens.",
      "highlights": [
        "Reduces KV cache needs by up to 75%",
        "Up to 6 times higher decoding throughput at contexts up to 1M tokens",
        "KDA kernel open-sourced in the flash-linear-attention library"
      ],
      "source_title": "moonshotai/Kimi-Linear-48B-A3B-Instruct (Hugging Face)",
      "source_url": "https://huggingface.co/moonshotai/Kimi-Linear-48B-A3B-Instruct",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/10#kimi-linear"
    },
    {
      "slug": "minimax-speech-2-6",
      "date": "2025-10-30",
      "model": "MiniMax Speech 2.6",
      "lab": "MiniMax",
      "lab_slug": "minimax",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "MiniMax Speech 2.6 is a text-to-speech model tuned for voice agents, with end-to-end latency under 250 milliseconds, direct reading of URLs, emails, phone numbers, dates, and amounts, and more natural prosody.",
      "highlights": [
        "Offered as speech-2.6-hd and speech-2.6-turbo",
        "Introduces Fluent LoRA for more fluent cloned voices",
        "Voice cloning across the 40+ languages the model supports"
      ],
      "source_title": "MiniMax Speech 2.6: The Ultimate Voice Agent Has Arrived",
      "source_url": "https://www.minimax.io/news/minimax-speech-26",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/10#minimax-speech-2-6"
    },
    {
      "slug": "gpt-oss-safeguard",
      "date": "2025-10-29",
      "model": "gpt-oss-safeguard-120b and gpt-oss-safeguard-20b",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "120b: 117B total, 5.1B active; 20b: 21B total, 3.6B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "gpt-oss-safeguard is a research preview of open-weight safety reasoning models, fine-tuned from gpt-oss, that classify messages, completions, and chats against a developer-provided policy interpreted at inference time.",
      "highlights": [
        "Released in two sizes, 120b and 20b, with weights on Hugging Face",
        "Same Apache 2.0 license as gpt-oss",
        "Provides chain-of-thought that developers can review to see how decisions were reached",
        "Developed with ROOST, which launched a model community alongside the release"
      ],
      "source_title": "Introducing gpt-oss-safeguard | OpenAI",
      "source_url": "https://openai.com/index/introducing-gpt-oss-safeguard/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/10#gpt-oss-safeguard"
    },
    {
      "slug": "amazon-nova-multimodal-embeddings",
      "date": "2025-10-28",
      "model": "Amazon Nova Multimodal Embeddings",
      "lab": "Amazon",
      "lab_slug": "amazon",
      "type": "embedding",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video",
        "audio"
      ],
      "output_modalities": [
        "embedding"
      ],
      "context_window": 8000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Amazon Nova Multimodal Embeddings is a single embedding model that maps text, documents, images, video and audio into one semantic space for cross-modal retrieval, agentic RAG and semantic search.",
      "highlights": [
        "Supports text, documents, images, video and audio inputs",
        "Four output dimension sizes trained with Matryoshka Representation Learning",
        "Up to 8K tokens of context, with text in up to 200 languages"
      ],
      "source_title": "Amazon Nova Multimodal Embeddings: State-of-the-art embedding model for agentic RAG and semantic search",
      "source_url": "https://aws.amazon.com/blogs/aws/amazon-nova-multimodal-embeddings-now-available-in-amazon-bedrock/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/10#amazon-nova-multimodal-embeddings"
    },
    {
      "slug": "granite-4-0-nano",
      "date": "2025-10-28",
      "model": "Granite 4.0 Nano",
      "lab": "IBM",
      "lab_slug": "ibm",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "350M to 1.6B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Smallest Granite 4.0 models, released in roughly 350M and 1B sizes as both hybrid H and conventional versions for edge and on-device use.",
      "highlights": [
        "1B models support 128K tokens; 350M models support 32K",
        "Base and instruct checkpoints released",
        "Released under Apache 2.0"
      ],
      "source_title": "ibm-granite/granite-4.0-h-1b · Hugging Face",
      "source_url": "https://huggingface.co/ibm-granite/granite-4.0-h-1b",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/10#granite-4-0-nano"
    },
    {
      "slug": "kaleido-14b-s2v",
      "date": "2025-10-28",
      "model": "Kaleido-14B-S2V",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": "14B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Kaleido is Z.ai's open-weights multi-subject reference video generation model, which generates video that keeps the identity of several reference subjects consistent. Z.ai released the Kaleido-14B-S2V checkpoints after open-sourcing the code.",
      "highlights": [
        "Training and inference code released October 22, 2025",
        "Checkpoints released October 28, 2025",
        "Weights under the MIT license"
      ],
      "source_title": "Kaleido (GitHub)",
      "source_url": "https://github.com/zai-org/Kaleido",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/10#kaleido-14b-s2v"
    },
    {
      "slug": "lfm2-colbert-350m",
      "date": "2025-10-28",
      "model": "LFM2-ColBERT-350M",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "embedding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "embedding"
      ],
      "context_window": null,
      "parameters": "353M",
      "model_license": "LFM Open License v1.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "A multilingual late-interaction retriever built on the LFM2 backbone that encodes queries and documents at the token level, supporting cross-lingual search where documents in one language are retrieved with queries in another.",
      "highlights": [
        "Document length 512 tokens and query length 32 tokens per model card",
        "Positioned as a drop-in replacement in existing RAG pipelines",
        "Available on Hugging Face with an interactive demo",
        "Succeeded by LFM2.5-ColBERT-350M in 2026"
      ],
      "source_title": "LFM2-ColBERT-350M: One Model to Embed Them All",
      "source_url": "https://www.liquid.ai/blog/lfm2-colbert-350m-one-model-to-embed-them-all",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/10#lfm2-colbert-350m"
    },
    {
      "slug": "minimax-hailuo-2-3",
      "date": "2025-10-28",
      "model": "MiniMax Hailuo 2.3",
      "lab": "MiniMax",
      "lab_slug": "minimax",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "MiniMax Hailuo 2.3 is a video generation model building on Hailuo 02, with improved body movement, stylization, and facial micro-expressions, and better response to motion instructions.",
      "highlights": [
        "Launched with Hailuo 2.3 Fast, a faster image-to-video variant",
        "Broader support for anime, illustration, ink wash, and game CG styles",
        "Available on the Hailuo AI website and app and the MiniMax API"
      ],
      "source_title": "MiniMax Hailuo 2.3: A New Level of Complex Video Performance & Media Agent",
      "source_url": "https://www.minimax.io/news/minimax-hailuo-23",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/10#minimax-hailuo-2-3"
    },
    {
      "slug": "nemotron-nano-2-vl",
      "date": "2025-10-28",
      "model": "Nemotron Nano 12B v2 VL",
      "lab": "NVIDIA",
      "lab_slug": "nvidia",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "12.6B",
      "model_license": "NVIDIA Open Model License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Multimodal document intelligence model that reads multiple document images such as invoices, receipts and manuals for question answering and extraction.",
      "highlights": [
        "128K tokens of combined input and output",
        "Handles multiple document images per request",
        "Released under the NVIDIA Open Model License"
      ],
      "source_title": "nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-BF16 · Hugging Face",
      "source_url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-12B-v2-VL-BF16",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/10#nemotron-nano-2-vl"
    },
    {
      "slug": "minimax-m2",
      "date": "2025-10-27",
      "model": "MiniMax M2",
      "lab": "MiniMax",
      "lab_slug": "minimax",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "MoE: 230B total, 10B active",
      "model_license": "Modified-MIT",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "MiniMax M2 is an open-weight mixture-of-experts model built for coding and agentic workflows, designed for end-to-end development in tools such as Claude Code and Cursor and for long chains of tool calls.",
      "highlights": [
        "230B total parameters with 10B active",
        "Weights on Hugging Face under a modified MIT license",
        "Uses interleaved thinking during tool use",
        "Open-sourced and launched on the MiniMax API the same day"
      ],
      "source_title": "MiniMax M2 & Agent: Ingenious in Simplicity",
      "source_url": "https://www.minimax.io/news/minimax-m2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/10#minimax-m2"
    },
    {
      "slug": "longcat-video",
      "date": "2025-10-25",
      "model": "LongCat-Video",
      "lab": "Meituan",
      "lab_slug": "meituan",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": "13.6B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "LongCat-Video is Meituan's 13.6B-parameter foundational video generation model, covering text-to-video, image-to-video, and video continuation in a single model, released with a technical report.",
      "highlights": [
        "13.6B parameters",
        "Single model for text-to-video, image-to-video, and video continuation",
        "Weights released under MIT"
      ],
      "source_title": "GitHub - meituan-longcat/LongCat-Video",
      "source_url": "https://github.com/meituan-longcat/LongCat-Video",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/10#longcat-video"
    },
    {
      "slug": "ltx-2",
      "date": "2025-10-23",
      "model": "LTX-2",
      "lab": "Lightricks",
      "lab_slug": "lightricks",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video",
        "audio"
      ],
      "output_modalities": [
        "video",
        "audio"
      ],
      "context_window": null,
      "parameters": "19B",
      "model_license": "LTX-2 Community License",
      "available_at_launch": [
        "Open weights",
        "API",
        "Consumer app"
      ],
      "summary": "Unified audio-video foundation model that generates synchronized video and sound in one process, with native 4K at up to 50 fps and clips up to 10 seconds at launch.",
      "highlights": [
        "Full open weights and code released January 6, 2026",
        "19B-parameter checkpoints with FP8 and FP4 variants",
        "Multi-keyframe conditioning and LoRA fine-tuning"
      ],
      "source_title": "Release Notes | LTX",
      "source_url": "https://ltx.io/release-notes",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/10#ltx-2"
    },
    {
      "slug": "seed3d-1-0",
      "date": "2025-10-23",
      "model": "Seed3D 1.0",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "image"
      ],
      "output_modalities": [
        "3d"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Seed3D 1.0 is a DiT-based 3D generation model that creates high-fidelity, simulation-ready 3D assets with detailed textures and materials from a single image, aimed at world simulators and embodied AI.",
      "highlights": [
        "Geometry generation model has 1.5B parameters",
        "Volcano Engine describes outputs as meshes with PBR materials",
        "API released alongside the technical report",
        "Extensible to scene generation"
      ],
      "source_title": "Seed3D 1.0 Released: Generate High-Fidelity 3D Models from Single Images, Featuring SOTA Texturing",
      "source_url": "https://seed.bytedance.com/en/blog/seed3d-1-0-released-generate-high-fidelity-3d-models-from-single-images-featuring-sota-texturing",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/10#seed3d-1-0"
    },
    {
      "slug": "hunyuanworld-mirror",
      "date": "2025-10-22",
      "model": "HunyuanWorld-Mirror (HunyuanWorld 1.1)",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "image",
        "video"
      ],
      "output_modalities": [
        "3d"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Tencent HunyuanWorld-Mirror Community License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "HunyuanWorld-Mirror, released as HunyuanWorld 1.1, is an open feed-forward model for 3D geometric prediction that creates 3D worlds from videos or multi-view images, optionally using camera pose, intrinsics, and depth priors.",
      "highlights": [
        "Outputs point clouds, multi-view depths, camera parameters, surface normals, and 3D Gaussians",
        "Training and evaluation code followed on November 7, 2025"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/HunyuanWorld-Mirror: [ICML 2026] WorldMirror: Fast and Universal 3D reconstruction model for versatile tasks",
      "source_url": "https://github.com/Tencent-Hunyuan/HunyuanWorld-Mirror",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/10#hunyuanworld-mirror"
    },
    {
      "slug": "lfm2-vl-3b",
      "date": "2025-10-22",
      "model": "LFM2-VL-3B",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32768,
      "parameters": "3B",
      "model_license": "LFM Open License v1.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "A 3B-parameter vision-language model built on the LFM2-2.6B backbone with a SigLIP2 400M NaFlex encoder, adding more capacity to the LFM2-VL family and extending visual understanding to ten languages.",
      "highlights": [
        "Processes images at native resolution with variable aspect ratios",
        "Adjustable vision tokens per image to balance speed and quality",
        "Available on Hugging Face under the LFM Open License and on LEAP"
      ],
      "source_title": "LFM2-VL-3B: A New Efficient Vision-Language for the Edge",
      "source_url": "https://www.liquid.ai/blog/lfm2-vl-3b-a-new-efficient-vision-language-for-the-edge",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/10#lfm2-vl-3b"
    },
    {
      "slug": "olmocr-2",
      "date": "2025-10-22",
      "model": "olmOCR 2",
      "lab": "Ai2",
      "lab_slug": "ai2",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "7B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Updated OCR model for English digitized print documents, built on Qwen2.5-VL-7B and trained with unit test rewards for document correctness.",
      "highlights": [
        "Released as olmOCR-2-7B-1025",
        "Fine-tuned on olmOCR-mix-1025 with 270,000 PDF pages",
        "Released under Apache 2.0"
      ],
      "source_title": "olmOCR 2: Unit test rewards for document OCR | Ai2",
      "source_url": "https://allenai.org/blog/olmocr-2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/10#olmocr-2"
    },
    {
      "slug": "deepseek-ocr",
      "date": "2025-10-20",
      "model": "DeepSeek-OCR",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "DeepSeek-OCR is an open-weights vision-language model that explores contexts optical compression, studying the role of vision encoders from an LLM-centric viewpoint. It converts document images to text and markdown.",
      "highlights": [
        "Released on GitHub and Hugging Face on October 20, 2025",
        "Supported in upstream vLLM from October 23, 2025",
        "Weights under the MIT License"
      ],
      "source_title": "DeepSeek-OCR: Contexts Optical Compression (GitHub)",
      "source_url": "https://github.com/deepseek-ai/DeepSeek-OCR",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/10#deepseek-ocr"
    },
    {
      "slug": "paddleocr-vl",
      "date": "2025-10-16",
      "model": "PaddleOCR-VL",
      "lab": "Baidu",
      "lab_slug": "baidu",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "image",
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "0.9B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "PaddleOCR-VL is Baidu's open document-parsing system built around PaddleOCR-VL-0.9B, a compact vision-language model that pairs a NaViT-style dynamic-resolution visual encoder with the ERNIE-4.5-0.3B language model.",
      "highlights": [
        "Supports 109 languages",
        "Recognizes text, tables, formulas, and charts, including handwriting and historical documents",
        "Uses PP-DocLayoutV2 for layout analysis and reading order, with Markdown and JSON output"
      ],
      "source_title": "PaddleOCR-VL: Boosting Multilingual Document Parsing via a 0.9B Ultra-Compact Vision-Language Model",
      "source_url": "https://ernie.baidu.com/research/posts/paddleocr-vl/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/10#paddleocr-vl"
    },
    {
      "slug": "claude-haiku-4-5",
      "date": "2025-10-15",
      "model": "Claude Haiku 4.5",
      "lab": "Anthropic",
      "lab_slug": "anthropic",
      "type": "small",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 200000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Cloud platforms"
      ],
      "summary": "Claude Haiku 4.5 is Anthropic's small model, which Anthropic says matches Claude Sonnet 4 on coding at lower cost and higher speed, aimed at real-time, low-latency work and multi-agent setups that run subtasks in parallel.",
      "highlights": [
        "API model ID claude-haiku-4-5 (snapshot claude-haiku-4-5-20251001)",
        "200K-token context window and 64K max output tokens",
        "Available in the Claude apps, Claude Code, the Claude API, Amazon Bedrock, and Google Cloud Vertex AI"
      ],
      "source_title": "Introducing Claude Haiku 4.5",
      "source_url": "https://www.anthropic.com/news/claude-haiku-4-5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/10#claude-haiku-4-5"
    },
    {
      "slug": "doubao-seed-1-6-lite",
      "date": "2025-10-15",
      "model": "Doubao-Seed-1.6-lite",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Volcano Engine upgraded Doubao-Seed-1.6 with four selectable thinking lengths (Minimal, Low, Medium, High) and released Doubao-Seed-1.6-lite, a smaller and faster deep-thinking model for enterprise workloads.",
      "highlights": [
        "Adds Minimal, Low, Medium, and High thinking-length settings",
        "Volcano Engine calls it the first model in China with native graded thinking-length control",
        "Doubao-Seed-1.6-lite released the same day"
      ],
      "source_title": "豆包大模型1.6升级：国内首个原生支持“分档调节思考长度”的大模型-发布中心-火山引擎",
      "source_url": "https://www.volcengine.com/news/detail/23",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/10#doubao-seed-1-6-lite"
    },
    {
      "slug": "doubao-seed-1-6-rerank",
      "date": "2025-10-15",
      "model": "Doubao-Seed-1.6-Rerank",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "reranker",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Doubao-Seed-1.6-Rerank is an all-modality reranking model that reorders retrieved text, image, and video results, integrated into Volcano Engine's VikingDB vector database alongside Doubao-Seed-1.6-Embedding for end-to-end retrieval.",
      "highlights": [
        "Reranks mixed text, image, and video candidates",
        "Understands negation and comparison logic in queries",
        "Produces stable relevance scores suited to threshold filtering"
      ],
      "source_title": "向量库支持全模态Doubao-Seed-1.6-Rerank模型，实现VectorDB+Embedding+Rerank全环节融合-发布中心-火山引擎",
      "source_url": "https://www.volcengine.com/news/detail/20",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/10#doubao-seed-1-6-rerank"
    },
    {
      "slug": "seed-tts-2-0",
      "date": "2025-10-15",
      "model": "Seed-TTS 2.0 and Seed-ICL 2.0",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Seed-TTS 2.0 and Seed-ICL 2.0 are Doubao's speech synthesis and voice cloning models, introducing a query-response conversational synthesis approach that uses dialogue context and natural-language instructions to control tone, emotion, and speaking rate.",
      "highlights": [
        "Takes the user's query and background information as context for delivery",
        "Speaking rate, emotion, and style can be steered with text instructions",
        "Available in the Doubao Speech console on Volcano Engine"
      ],
      "source_title": "从文本朗读到情感演绎，豆包语音模型重磅升级-发布中心-火山引擎",
      "source_url": "https://www.volcengine.com/news/detail/14",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/10#seed-tts-2-0"
    },
    {
      "slug": "veo-3-1",
      "date": "2025-10-15",
      "model": "Veo 3.1",
      "lab": "Google",
      "lab_slug": "google",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video",
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API",
        "Cloud platforms"
      ],
      "summary": "Veo 3.1 is Google DeepMind's upgrade to Veo 3, bringing richer generated audio, stronger prompt adherence for image-to-video and more narrative control. It launched with a Veo 3.1 Fast variant across Google's apps and APIs.",
      "highlights": [
        "Flow features Ingredients to Video, Frames to Video and Extend now include audio",
        "Veo 3.1 and Veo 3.1 Fast entered public preview in the Gemini API on launch day",
        "Generally available on Vertex AI from November 17, 2025",
        "Also available in the Gemini app"
      ],
      "source_title": "Introducing Veo 3.1 and advanced capabilities in Flow",
      "source_url": "https://blog.google/technology/ai/veo-updates-flow/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/10#veo-3-1"
    },
    {
      "slug": "mai-image-1",
      "date": "2025-10-13",
      "model": "MAI-Image-1",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "MAI-Image-1 is Microsoft AI's first image generation model developed entirely in-house. It later became available in Bing Image Creator and in Copilot Audio Expressions Story Mode.",
      "highlights": [
        "Microsoft's first in-house text-to-image model",
        "Added to Bing Image Creator and Copilot on November 4, 2025",
        "Available in countries that can access Bing Image Creator and Copilot Labs"
      ],
      "source_title": "Introducing MAI-Image-1, debuting in the top 10 on LMArena",
      "source_url": "https://microsoft.ai/news/introducing-mai-image-1-debuting-in-the-top-10-on-lmarena/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/10#mai-image-1"
    },
    {
      "slug": "jamba-reasoning-3b",
      "date": "2025-10-08",
      "model": "Jamba Reasoning 3B",
      "lab": "AI21 Labs",
      "lab_slug": "ai21",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "3B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Compact open-source reasoning model built on AI21's hybrid SSM-Transformer architecture, designed to run on phones and computers for long-context reasoning.",
      "highlights": [
        "Handles 256K tokens and can stretch to 1M",
        "Runs locally via llama.cpp or LM Studio",
        "Available on Hugging Face and Kaggle"
      ],
      "source_title": "Introducing Jamba Reasoning 3B: Tiny Model, Huge Possibilities",
      "source_url": "https://www.ai21.com/blog/introducing-jamba-reasoning-3b/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/10#jamba-reasoning-3b"
    },
    {
      "slug": "gemini-2-5-computer-use",
      "date": "2025-10-07",
      "model": "Gemini 2.5 Computer Use",
      "lab": "Google",
      "lab_slug": "google",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "image",
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms"
      ],
      "summary": "The Gemini 2.5 Computer Use model is a specialized model built on Gemini 2.5 Pro's visual understanding and reasoning that powers agents able to interact with user interfaces in browsers and on mobile.",
      "highlights": [
        "Public preview in the Gemini API on Google AI Studio and Vertex AI",
        "Model page lists a 128,000 input token limit and 64,000 output tokens",
        "Computer Use later became a tool in Gemini 3 and 3.5 models, reaching general availability on Oct 30, 2026"
      ],
      "source_title": "Introducing the Gemini 2.5 Computer Use model",
      "source_url": "https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-computer-use-model/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/10#gemini-2-5-computer-use"
    },
    {
      "slug": "lfm2-8b-a1b",
      "date": "2025-10-07",
      "model": "LFM2-8B-A1B",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32768,
      "parameters": "MoE: 8.3B total, 1.5B active",
      "model_license": "LFM Open License v1.0",
      "available_at_launch": [
        "Consumer app",
        "Open weights"
      ],
      "summary": "Liquid AI's first on-device mixture-of-experts model, an early checkpoint that adds sparse MoE feed-forward layers to the LFM2 backbone, with 8.3B total and 1.5B active parameters per token.",
      "highlights": [
        "32 experts per MoE block with top-4 routing; 18 convolution and 6 GQA blocks",
        "Pre-trained on about 12T tokens with more code than LFM2-2.6B",
        "Weights and GGUF quants on Hugging Face; testable on Liquid Playground",
        "Later replaced by LFM2.5-8B-A1B (May 2026)"
      ],
      "source_title": "LFM2-8B-A1B: An Efficient On-device Mixture-of-Experts",
      "source_url": "https://www.liquid.ai/blog/lfm2-8b-a1b-an-efficient-on-device-mixture-of-experts",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/10#lfm2-8b-a1b"
    },
    {
      "slug": "gpt-image-1-mini",
      "date": "2025-10-06",
      "model": "GPT Image 1 Mini",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "GPT Image 1 Mini is a cost-efficient version of GPT Image 1 for image generation and editing, released in the API at OpenAI DevDay 2025 alongside new audio and video models.",
      "highlights": [
        "API model ID gpt-image-1-mini on the image generation and edit endpoints",
        "Released at OpenAI DevDay on October 6, 2025",
        "Accepts text and image input and outputs images, per the API model page"
      ],
      "source_title": "Changelog | OpenAI API",
      "source_url": "https://developers.openai.com/api/docs/changelog",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/10#gpt-image-1-mini"
    },
    {
      "slug": "gpt-realtime-mini",
      "date": "2025-10-06",
      "model": "gpt-realtime-mini and gpt-audio-mini",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "audio",
        "image"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "gpt-realtime-mini and gpt-audio-mini are cost-efficient versions of gpt-realtime and gpt-audio for speech-to-speech applications, released at OpenAI DevDay 2025 for the Realtime API and Chat Completions API respectively.",
      "highlights": [
        "gpt-realtime-mini runs in the Realtime API; gpt-audio-mini in Chat Completions",
        "Context: 32,000 tokens for gpt-realtime-mini and 128,000 for gpt-audio-mini",
        "Image input is supported by gpt-realtime-mini only, per the API model pages",
        "Updated snapshots dated 2025-12-15 followed in December 2025"
      ],
      "source_title": "Changelog | OpenAI API",
      "source_url": "https://developers.openai.com/api/docs/changelog",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/10#gpt-realtime-mini"
    },
    {
      "slug": "granite-4-0",
      "date": "2025-10-02",
      "model": "Granite 4.0",
      "lab": "IBM",
      "lab_slug": "ibm",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "H-Small 32B (9B active), H-Tiny 7B (1B active), H-Micro 3B, Micro 3B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Fourth-generation Granite language models that add hybrid H variants, including mixture-of-experts models, alongside a conventional dense Micro model, all with 128K-token context.",
      "highlights": [
        "H-Small activates 9B of 32B parameters",
        "Base and instruct checkpoints released",
        "Released under Apache 2.0"
      ],
      "source_title": "ibm-granite/granite-4.0-h-small · Hugging Face",
      "source_url": "https://huggingface.co/ibm-granite/granite-4.0-h-small",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/10#granite-4-0"
    },
    {
      "slug": "lfm2-audio-1-5b",
      "date": "2025-10-01",
      "model": "LFM2-Audio-1.5B",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": 32768,
      "parameters": "1.5B",
      "model_license": "LFM Open License v1.0",
      "available_at_launch": [
        "Consumer app",
        "Open weights"
      ],
      "summary": "An end-to-end audio and text foundation model that extends LFM2-1.2B to accept and generate both speech and text, covering conversational chat, speech recognition, text-to-speech, and audio classification in one 1.5B-parameter model.",
      "highlights": [
        "Continuous embeddings for audio input; discrete audio tokens for output",
        "Tokenizer-free input: raw waveform chunked into segments of about 80 ms",
        "Supported language per model card: English",
        "Available on Hugging Face and Liquid Playground"
      ],
      "source_title": "LFM2-Audio: An End-to-End Audio Foundation Model",
      "source_url": "https://www.liquid.ai/blog/lfm2-audio-an-end-to-end-audio-foundation-model",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/10#lfm2-audio-1-5b"
    },
    {
      "slug": "glm-4-6",
      "date": "2025-09-30",
      "model": "GLM-4.6",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 200000,
      "parameters": "MoE: 355B total, 32B active",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "GLM-4.6 is Z.ai's flagship coding model and the successor to GLM-4.5. It expands the context window from 128K to 200K tokens and supports tool use during reasoning.",
      "highlights": [
        "Better real-world performance in Claude Code, Cline, Roo Code, and Kilo Code",
        "Stronger tool-using and search-based agents",
        "Up to 128K output tokens; weights under the MIT license"
      ],
      "source_title": "GLM-4.6: Advanced Agentic, Reasoning and Coding Capabilities",
      "source_url": "https://z.ai/blog/glm-4.6",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#glm-4-6"
    },
    {
      "slug": "sora-2",
      "date": "2025-09-30",
      "model": "Sora 2",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video",
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Sora 2 is OpenAI's flagship video and audio generation model, which OpenAI says is more physically accurate and controllable than prior systems and adds synchronized dialogue and sound effects. It launched with a new Sora social app.",
      "highlights": [
        "Initial invite-based rollout in the U.S. and Canada via the Sora iOS app and sora.com",
        "ChatGPT Pro users could use the higher-quality Sora 2 Pro model on sora.com",
        "API models sora-2 and sora-2-pro released at DevDay on October 6, 2025"
      ],
      "source_title": "Sora 2 is here | OpenAI",
      "source_url": "https://openai.com/index/sora-2/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#sora-2"
    },
    {
      "slug": "claude-sonnet-4-5",
      "date": "2025-09-29",
      "model": "Claude Sonnet 4.5",
      "lab": "Anthropic",
      "lab_slug": "anthropic",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 200000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Cloud platforms"
      ],
      "summary": "Claude Sonnet 4.5 is an Anthropic model aimed at coding, building complex agents, and computer use, with gains in reasoning and math. Anthropic released the Claude Agent SDK and API context editing and memory tools alongside it.",
      "highlights": [
        "API model ID claude-sonnet-4-5 (snapshot claude-sonnet-4-5-20250929)",
        "200K-token context window and 64K max output tokens",
        "Deprecated September 30, 2026, with Claude API retirement set for November 30, 2026"
      ],
      "source_title": "Introducing Claude Sonnet 4.5",
      "source_url": "https://www.anthropic.com/news/claude-sonnet-4-5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#claude-sonnet-4-5"
    },
    {
      "slug": "deepseek-v3-2-exp",
      "date": "2025-09-29",
      "model": "DeepSeek-V3.2-Exp",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "DeepSeek-V3.2-Exp is an experimental model built on V3.1-Terminus that debuts DeepSeek Sparse Attention for more efficient long-context training and inference. DeepSeek says it performs on par with V3.1-Terminus.",
      "highlights": [
        "Live on the DeepSeek app, web, and API at launch",
        "V3.1-Terminus stayed available on a temporary API for comparison until October 15, 2025",
        "Released with GPU kernels in TileLang and CUDA and a technical report"
      ],
      "source_title": "Introducing DeepSeek-V3.2-Exp",
      "source_url": "https://api-docs.deepseek.com/news/news250929",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#deepseek-v3-2-exp"
    },
    {
      "slug": "hunyuanimage-3-0",
      "date": "2025-09-28",
      "model": "HunyuanImage 3.0",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "MoE: 80B total, 13B active",
      "model_license": "Tencent Hunyuan Community License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "HunyuanImage 3.0 is an open native multimodal image generation model that Tencent describes as the largest open-source image generation mixture-of-experts model at release.",
      "highlights": [
        "64 experts with 80B total parameters and 13B activated per token",
        "Technical report published the same day",
        "vLLM-accelerated inference added October 30, 2025"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/HunyuanImage-3.0: HunyuanImage-3.0: A Powerful Native Multimodal Model for Image Generation",
      "source_url": "https://github.com/Tencent-Hunyuan/HunyuanImage-3.0",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#hunyuanimage-3-0"
    },
    {
      "slug": "hunyuan3d-omni",
      "date": "2025-09-26",
      "model": "Hunyuan3D-Omni",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "image"
      ],
      "output_modalities": [
        "3d"
      ],
      "context_window": null,
      "parameters": "3.3B",
      "model_license": "Tencent Hunyuan 3D Omni Community License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Hunyuan3D-Omni is an open framework for controllable 3D asset generation that inherits the Hunyuan3D 2.1 structure and adds a unified control encoder for point cloud, voxel, skeleton, and bounding box conditions.",
      "highlights": [
        "3.3B-parameter image-to-shape model with multi-modal control",
        "Requires about 10 GB of VRAM for generation"
      ],
      "source_title": "License.txt · tencent/Hunyuan3D-Omni at main",
      "source_url": "https://huggingface.co/tencent/Hunyuan3D-Omni/blob/main/License.txt",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#hunyuan3d-omni"
    },
    {
      "slug": "gemini-robotics-er-1-5",
      "date": "2025-09-25",
      "model": "Gemini Robotics-ER 1.5",
      "lab": "Google",
      "lab_slug": "google",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1048576,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Gemini Robotics-ER 1.5 is an embodied reasoning model that acts as a high-level planner for robots, and the first Gemini Robotics model made broadly available to developers.",
      "highlights": [
        "Preview in Google AI Studio and the Gemini API",
        "A thinking model with a flexible thinking budget to trade latency for accuracy",
        "Replaced by Gemini Robotics-ER 1.6 and shut down on Apr 30, 2026"
      ],
      "source_title": "Building the Next Generation of Physical Agents with Gemini Robotics-ER 1.5",
      "source_url": "https://developers.googleblog.com/building-the-next-generation-of-physical-agents-with-gemini-robotics-er-15/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#gemini-robotics-er-1-5"
    },
    {
      "slug": "lfm2-liquid-nanos",
      "date": "2025-09-25",
      "model": "Liquid Nanos (LFM2-Extract, LFM2-350M-ENJP-MT, LFM2-1.2B-RAG, LFM2-1.2B-Tool, LFM2-350M-Math)",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "350M and 1.2B checkpoints",
      "model_license": "LFM Open License v1.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "A first set of experimental task-specific LFM2 models for on-device agentic building blocks: data extraction, English-Japanese translation, retrieval-augmented question answering, tool calling, and math reasoning, released as open-weight 350M and 1.2B checkpoints.",
      "highlights": [
        "Models: LFM2-350M-Extract, LFM2-1.2B-Extract, LFM2-350M-ENJP-MT, LFM2-1.2B-RAG, LFM2-1.2B-Tool, LFM2-350M-Math",
        "Extract models turn unstructured documents into JSON, XML, or YAML",
        "Available on Hugging Face and Liquid's LEAP platform for iOS, Android, and laptops"
      ],
      "source_title": "Introducing Liquid Nanos: frontier-grade performance on everyday devices",
      "source_url": "https://www.liquid.ai/blog/introducing-liquid-nanos-frontier-grade-performance-on-everyday-devices",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#lfm2-liquid-nanos"
    },
    {
      "slug": "meta-cwm",
      "date": "2025-09-24",
      "model": "Code World Model (CWM)",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 131072,
      "parameters": "32B",
      "model_license": "FAIR Non-Commercial Research License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Code World Model is a 32B dense, decoder-only open-weights LLM from Meta FAIR, released to support research on code generation with world models.",
      "highlights": [
        "Dense decoder-only model trained with up to 131k tokens of context",
        "Checkpoints released after mid-training, SFT and RL",
        "Intended for non-commercial research, not as a chat assistant"
      ],
      "source_title": "CWM: An Open-Weights LLM for Research on Code Generation with World Models",
      "source_url": "https://ai.meta.com/research/publications/cwm-an-open-weights-llm-for-research-on-code-generation-with-world-models/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#meta-cwm"
    },
    {
      "slug": "qwen3-max",
      "date": "2025-09-24",
      "model": "Qwen3-Max",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "MoE: over 1T total",
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Qwen3-Max is Alibaba's largest Qwen model at launch, with more than 1 trillion parameters pre-trained on 36T tokens. The official release followed a preview and improved coding and agent capabilities.",
      "highlights": [
        "Qwen3-Max-Instruct available via the Alibaba Cloud API and Qwen Chat",
        "Trained with a 1M-token context using the ChunkFlow strategy",
        "A Thinking variant was still in training at launch"
      ],
      "source_title": "Qwen3-Max: Just Scale it",
      "source_url": "https://qwen.ai/blog?id=qwen3-max",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#qwen3-max"
    },
    {
      "slug": "lfm2-2-6b",
      "date": "2025-09-23",
      "model": "LFM2-2.6B",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32768,
      "parameters": "2.6B",
      "model_license": "LFM Open License v1.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "The largest dense model in the LFM2 series at its release, a 2.6B-parameter hybrid convolution and grouped query attention model trained on 10 trillion tokens and tuned for English and Japanese.",
      "highlights": [
        "Also supports French, Spanish, German, Italian, Portuguese, Arabic, Chinese, and Korean",
        "32,768-token context window per model card",
        "Released on Hugging Face under the LFM Open License"
      ],
      "source_title": "Introducing LFM2-2.6B: Redefining Efficiency in Language Models",
      "source_url": "https://www.liquid.ai/blog/introducing-lfm2-2-6b-redefining-efficiency-in-language-models",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#lfm2-2-6b"
    },
    {
      "slug": "qwen-image-edit-2509",
      "date": "2025-09-23",
      "model": "Qwen-Image-Edit-2509",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "20B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Consumer app",
        "Open weights"
      ],
      "summary": "Qwen-Image-Edit-2509 is the September 2025 monthly iteration of the open-weights Qwen-Image-Edit model, adding support for multi-image input and improving consistency in edits compared with the August release.",
      "highlights": [
        "Multi-image editing support",
        "Available through the Image Editing feature on Qwen Chat",
        "Weights under the Apache 2.0 license"
      ],
      "source_title": "Qwen-Image-Edit-2509: Multi-Image Support, Improved Consistency",
      "source_url": "https://qwen.ai/blog?id=qwen-image-edit-2509",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#qwen-image-edit-2509"
    },
    {
      "slug": "qwen3-livetranslate-flash",
      "date": "2025-09-23",
      "model": "Qwen3-LiveTranslate-Flash",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio",
        "video"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Qwen3-LiveTranslate-Flash is Qwen's real-time multilingual audio and video interpretation model built on Qwen3-Omni. It supports offline and live translation in 18 languages and uses visual context to improve translations.",
      "highlights": [
        "First Qwen interpretation model with visual context augmentation",
        "Trained on millions of hours of multimodal data",
        "Later followed by Qwen3.5-LiveTranslate-Flash and Qwen3.8-LiveTranslate"
      ],
      "source_title": "Qwen3-LiveTranslate: Real-Time Multimodal Interpretation",
      "source_url": "https://qwen.ai/blog?id=qwen3-livetranslate",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#qwen3-livetranslate-flash"
    },
    {
      "slug": "qwen3-vl",
      "date": "2025-09-23",
      "model": "Qwen3-VL",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "MoE: 235B total, 22B active (flagship)",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "Qwen3-VL is Qwen's vision-language series, opened with Qwen3-VL-235B-A22B in Instruct and Thinking versions. Qwen says its text performance matches the flagship Qwen3-235B-A22B-2507 language model.",
      "highlights": [
        "Native 256K-token context, expandable to 1M tokens",
        "OCR expanded to 32 languages from 10",
        "30B-A3B followed October 4, 4B and 8B on October 15, and 2B and 32B on October 21, 2025"
      ],
      "source_title": "Qwen3-VL: Sharper Vision, Deeper Thought, Broader Action",
      "source_url": "https://qwen.ai/blog?id=qwen3-vl",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#qwen3-vl"
    },
    {
      "slug": "qwen3guard",
      "date": "2025-09-23",
      "model": "Qwen3Guard",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "0.6B, 4B, 8B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Qwen3Guard is the first safety guardrail model series in the Qwen family, built on Qwen3 and fine-tuned for safety classification. It includes Qwen3Guard-Gen and Qwen3Guard-Stream, which detects unsafe content in real time during generation.",
      "highlights": [
        "Both variants come in 0.6B, 4B, and 8B sizes",
        "Stream variant enables token-level streaming safety detection",
        "Weights under the Apache 2.0 license"
      ],
      "source_title": "Qwen3Guard: Real-time Safety for Your Token Stream",
      "source_url": "https://qwen.ai/blog?id=qwen3guard",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#qwen3guard"
    },
    {
      "slug": "deepseek-v3-1-terminus",
      "date": "2025-09-22",
      "model": "DeepSeek-V3.1-Terminus",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "DeepSeek-V3.1-Terminus is an update to DeepSeek-V3.1 that addresses user feedback on language consistency and strengthens its code agent and search agent performance. It replaced V3.1 across the app, web, and API.",
      "highlights": [
        "Fewer Chinese and English mix-ups and fewer abnormal characters",
        "Upgraded both deepseek-chat (non-thinking) and deepseek-reasoner (thinking)",
        "Weights on Hugging Face"
      ],
      "source_title": "DeepSeek-V3.1-Terminus",
      "source_url": "https://api-docs.deepseek.com/news/news250922",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#deepseek-v3-1-terminus"
    },
    {
      "slug": "longcat-flash-thinking",
      "date": "2025-09-22",
      "model": "LongCat-Flash-Thinking",
      "lab": "Meituan",
      "lab_slug": "meituan",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "MoE: 560B total, about 27B active",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "LongCat-Flash-Thinking is Meituan's large reasoning model built on the LongCat-Flash mixture-of-experts architecture, released and open-sourced the same day with chat and API access.",
      "highlights": [
        "560B total parameters, sharing the LongCat-Flash architecture",
        "Available free in LongCat Chat and via API as model=LongCat-Flash-Thinking",
        "Weights on Hugging Face and GitHub under MIT"
      ],
      "source_title": "Change Log | LongCat API Docs",
      "source_url": "https://longcat.chat/platform/docs/change-log",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#longcat-flash-thinking"
    },
    {
      "slug": "qianfan-vl",
      "date": "2025-09-22",
      "model": "Qianfan-VL",
      "lab": "Baidu",
      "lab_slug": "baidu",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "image",
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32768,
      "parameters": "3B, 8B, 70B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Qianfan-VL is a Baidu AI Cloud family of vision-language models in 3B, 8B, and 70B sizes, continuously pre-trained for enterprise OCR and document understanding, with chain-of-thought reasoning in the 8B and 70B versions.",
      "highlights": [
        "3B model built on Qwen2.5-3B; 8B and 70B models based on the Llama 3.1 architecture",
        "InternViT-based vision encoder with dynamic patching up to 4K resolution",
        "32K context across all three sizes"
      ],
      "source_title": "GitHub - baidubce/Qianfan-VL: Qianfan-VL: Domain-Enhanced Universal Vision-Language Models",
      "source_url": "https://github.com/baidubce/Qianfan-VL",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#qianfan-vl"
    },
    {
      "slug": "qwen3-omni",
      "date": "2025-09-22",
      "model": "Qwen3-Omni",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": null,
      "parameters": "MoE: 30B total, 3B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "Qwen3-Omni is Qwen's natively end-to-end multilingual omni-modal model, processing text, images, audio, and video and responding in real time with text and natural speech. It was released as Instruct, Thinking, and Captioner models.",
      "highlights": [
        "Text interaction in 119 languages, speech understanding in 19, and speech generation in 10",
        "Qwen3-Omni-30B-A3B-Captioner is an open audio captioning model",
        "Weights under the Apache 2.0 license"
      ],
      "source_title": "Qwen3-Omni: Natively Omni-Modal Foundation Models!",
      "source_url": "https://qwen.ai/blog?id=qwen3-omni",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#qwen3-omni"
    },
    {
      "slug": "qwen3-tts-flash",
      "date": "2025-09-22",
      "model": "Qwen3-TTS-Flash",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Qwen3-TTS-Flash is Qwen's flagship text-to-speech model on the Qwen API, supporting multi-timbre, multilingual, and multi-dialect speech synthesis. It aims to produce natural and expressive speech.",
      "highlights": [
        "17 voices, each supporting 10 languages",
        "Dialects including Hokkien, Wu, Cantonese, and Sichuanese",
        "Available through the Qwen API on Alibaba Cloud Model Studio"
      ],
      "source_title": "Qwen3-TTS-Flash: Multi-timbre & Multi-lingual & Multi-dialect Speech Synthesis",
      "source_url": "https://qwen.ai/blog?id=qwen3-tts",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#qwen3-tts-flash"
    },
    {
      "slug": "grok-4-fast",
      "date": "2025-09-19",
      "model": "Grok 4 Fast",
      "lab": "xAI",
      "lab_slug": "xai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 2000000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Grok 4 Fast is xAI's cost-efficient model built on learnings from Grok 4, with a unified architecture in which the same weights handle reasoning and non-reasoning modes, plus web and X search.",
      "highlights": [
        "API model IDs: grok-4-fast-reasoning and grok-4-fast-non-reasoning, each with 2M-token context",
        "Available to all grok.com and app users, including free users",
        "Also offered on OpenRouter and Vercel AI Gateway"
      ],
      "source_title": "Grok 4 Fast | SpaceXSI",
      "source_url": "https://x.ai/news/grok-4-fast",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#grok-4-fast"
    },
    {
      "slug": "kling-2-5-turbo",
      "date": "2025-09-19",
      "model": "Kling 2.5 Turbo",
      "lab": "Kuaishou",
      "lab_slug": "kuaishou",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Kling 2.5 Turbo is an upgraded Kling video model with improved text-to-video and image-to-video quality, including better prompt adherence, motion, style consistency, and aesthetics.",
      "highlights": [
        "Improved handling of multi-step instructions and causal relationships in prompts",
        "Larger-amplitude motion and camera movement with more realistic physics",
        "Kuaishou's press release dates the launch to September 23, 2025"
      ],
      "source_title": "Kling AI Release Notes & Version History | Kling AI",
      "source_url": "https://kling.ai/release-note/release-history",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#kling-2-5-turbo"
    },
    {
      "slug": "mimo-audio",
      "date": "2025-09-19",
      "model": "MiMo-Audio",
      "lab": "Xiaomi",
      "lab_slug": "xiaomi",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": null,
      "parameters": "7B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "MiMo-Audio is Xiaomi's open-source audio language model, pretrained on more than 100 million hours of audio and showing few-shot learning on audio tasks such as voice conversion, style transfer, and speech continuation.",
      "highlights": [
        "Released as MiMo-Audio-7B-Base and MiMo-Audio-7B-Instruct",
        "Includes MiMo-Audio-Tokenizer, a 1.2B-parameter 25 Hz tokenizer",
        "Instruct model adds thinking to audio understanding and generation",
        "Technical report dated September 19, 2025 on the MiMo site"
      ],
      "source_title": "XiaomiMiMo/MiMo-Audio-7B-Instruct",
      "source_url": "https://huggingface.co/XiaomiMiMo/MiMo-Audio-7B-Instruct",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#mimo-audio"
    },
    {
      "slug": "wan2-2-animate",
      "date": "2025-09-19",
      "model": "Wan2.2-Animate-14B",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": "14B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Wan2.2-Animate-14B is a unified model for character animation and replacement that replicates full-body movement and facial expression from a reference video.",
      "highlights": [
        "Model weights and inference code released",
        "Diffusers integration added November 13, 2025",
        "Released under the Apache 2.0 license"
      ],
      "source_title": "Wan2.2 (GitHub)",
      "source_url": "https://github.com/Wan-Video/Wan2.2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#wan2-2-animate"
    },
    {
      "slug": "luma-ray3",
      "date": "2025-09-18",
      "model": "Ray3",
      "lab": "Luma AI",
      "lab_slug": "luma-ai",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Video model that Luma describes as the first reasoning video model, able to evaluate and refine its own outputs, and the first to generate 10, 12 and 16-bit HDR video in ACES2065-1 EXR format.",
      "highlights": [
        "Native HDR EXR generation",
        "Draft Mode for faster exploration",
        "Launched in Dream Machine with Adobe Firefly as first outside partner"
      ],
      "source_title": "Luma AI launches Ray3",
      "source_url": "https://lumalabs.ai/news/ray3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#luma-ray3"
    },
    {
      "slug": "granite-docling",
      "date": "2025-09-17",
      "model": "Granite-Docling 258M",
      "lab": "IBM",
      "lab_slug": "ibm",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "258M",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Compact document conversion vision-language model built on the Idefics3 architecture with a SigLIP2 vision encoder and a Granite 165M language model.",
      "highlights": [
        "Builds on the Idefics3 architecture",
        "Replaces the vision encoder with siglip2-base-patch16-512",
        "Released under Apache 2.0"
      ],
      "source_title": "ibm-granite/granite-docling-258M · Hugging Face",
      "source_url": "https://huggingface.co/ibm-granite/granite-docling-258M",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#granite-docling"
    },
    {
      "slug": "magistral-medium-1-2",
      "date": "2025-09-17",
      "model": "Magistral Medium 1.2",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "September 2025 update of Magistral Medium that turns it into a multimodal reasoning model and extends its context window to 128K tokens.",
      "highlights": [
        "API model ID magistral-medium-2509",
        "Mistral's model card lists September 18, 2025",
        "Deprecated May 22, 2026 in favor of Mistral Medium 3.5"
      ],
      "source_title": "Changelog | Mistral Docs",
      "source_url": "https://docs.mistral.ai/resources/changelogs",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#magistral-medium-1-2"
    },
    {
      "slug": "magistral-small-1-2",
      "date": "2025-09-17",
      "model": "Magistral Small 1.2",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "24B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "September 2025 update of the open Magistral Small reasoning model that adds image input and raises the context window from 40K to 128K tokens.",
      "highlights": [
        "API model ID magistral-small-2509",
        "Mistral's model card lists September 18, 2025",
        "Deprecated April 30, 2026 in favor of Mistral Small 4"
      ],
      "source_title": "Changelog | Mistral Docs",
      "source_url": "https://docs.mistral.ai/resources/changelogs",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#magistral-small-1-2"
    },
    {
      "slug": "reka-speech",
      "date": "2025-09-16",
      "model": "Reka Speech",
      "lab": "Reka",
      "lab_slug": "reka",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "850M (550M decoder, 300M audio encoder)",
      "model_license": null,
      "available_at_launch": [
        "Enterprise"
      ],
      "summary": "An 850M-parameter multilingual speech model for transcription and translation with timestamps, aimed at high-volume offline audio jobs and using a QK-cache alignment method on a modified vLLM for fast timestamping.",
      "highlights": [
        "Understands English, Japanese, Korean, Chinese, French, Spanish, German, Italian, Portuguese and Arabic",
        "Translates from English to each of the other supported languages and vice versa",
        "Available to customers by contacting Reka"
      ],
      "source_title": "Reka Speech: High Throughput Speech Transcription and Translation Model with Timestamps",
      "source_url": "https://reka.ai/news/reka-speech-high-throughput-speech-transcription-and-translation-model-with-timestamps",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#reka-speech"
    },
    {
      "slug": "gpt-5-codex",
      "date": "2025-09-15",
      "model": "GPT-5-Codex",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "coding",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 400000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "Enterprise"
      ],
      "summary": "GPT-5-Codex is a version of GPT-5 further optimized for agentic coding in Codex, trained on real-world software engineering work for both quick interactive sessions and long independent tasks, including code review.",
      "highlights": [
        "Default for Codex cloud tasks and code review; selectable in Codex CLI and the IDE extension",
        "Included with ChatGPT Plus, Pro, Business, Edu, and Enterprise plans",
        "Added to the Responses API as gpt-5-codex on September 23, 2025",
        "400,000-token context window per the API model page"
      ],
      "source_title": "Introducing upgrades to Codex | OpenAI",
      "source_url": "https://openai.com/index/introducing-upgrades-to-codex/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#gpt-5-codex"
    },
    {
      "slug": "vaultgemma",
      "date": "2025-09-12",
      "model": "VaultGemma",
      "lab": "Google",
      "lab_slug": "google",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1024,
      "parameters": "1B",
      "model_license": "Gemma Terms of Use",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "VaultGemma is a 1B-parameter Gemma 2-based model that Google Research trained from scratch with differential privacy. Google describes it as the largest open model trained this way, with a formal sequence-level privacy guarantee.",
      "highlights": [
        "Sequence-level DP guarantee of epsilon at most 2.0 and delta at most 1.1e-10, over 1,024-token sequences",
        "Pre-trained checkpoint only",
        "Weights on Hugging Face and Kaggle with a technical report"
      ],
      "source_title": "VaultGemma: The world's most capable differentially private LLM",
      "source_url": "https://research.google/blog/vaultgemma-the-worlds-most-capable-differentially-private-llm/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#vaultgemma"
    },
    {
      "slug": "qwen3-next",
      "date": "2025-09-11",
      "model": "Qwen3-Next",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 262144,
      "parameters": "MoE: 80B total, 3B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "Qwen3-Next is a new Qwen architecture for training and inference efficiency, launched with Qwen3-Next-80B-A3B in Instruct and Thinking versions. It combines hybrid attention with an ultra-sparse mixture-of-experts design.",
      "highlights": [
        "Base model matches dense Qwen3-32B at under 10% of its training cost, per Qwen",
        "More than 10 times higher throughput at contexts over 32K tokens",
        "Native 262,144-token context, extensible to 1,010,000 tokens"
      ],
      "source_title": "Qwen3-Next: Towards Ultimate Training & Inference Efficiency",
      "source_url": "https://qwen.ai/blog?id=qwen3-next",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#qwen3-next"
    },
    {
      "slug": "stable-audio-2-5",
      "date": "2025-09-10",
      "model": "Stable Audio 2.5",
      "lab": "Stability AI",
      "lab_slug": "stability-ai",
      "type": "music",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Cloud platforms",
        "Enterprise"
      ],
      "summary": "Enterprise-focused audio model with improved musical composition that generates tracks up to three minutes long in under two seconds on a GPU and supports audio inpainting.",
      "highlights": [
        "Post-trained with the Adversarial Relativistic-Contrastive method",
        "Trained on a fully licensed dataset",
        "Available via the Stability API, fal, Replicate, ComfyUI and on-premises licensing"
      ],
      "source_title": "Stability AI Introduces Stable Audio 2.5, the First Audio Model Built for Enterprise Sound Production at Scale",
      "source_url": "https://stability.ai/news-updates/stability-ai-introduces-stable-audio-25-the-first-audio-model-built-for-enterprise-sound-production-at-scale",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#stable-audio-2-5"
    },
    {
      "slug": "ernie-x1-1",
      "date": "2025-09-09",
      "model": "ERNIE X1.1",
      "lab": "Baidu",
      "lab_slug": "baidu",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "ERNIE X1.1 upgrades Baidu's ERNIE X1 deep-thinking model, using an iterative hybrid reinforcement learning framework to improve factuality, instruction following, and agentic task performance.",
      "highlights": [
        "Announced at WAVE SUMMIT 2025",
        "Available on the ERNIE Bot website, the Wenxiaoyan app, and Baidu AI Cloud Qianfan",
        "Like ERNIE X1, trained from ERNIE 4.5"
      ],
      "source_title": "文心大模型X1.1，正式发布！",
      "source_url": "https://home.baidu.com/home/index/news_detail/id/18038",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#ernie-x1-1"
    },
    {
      "slug": "ernie-4-5-21b-a3b-thinking",
      "date": "2025-09-09",
      "model": "ERNIE-4.5-21B-A3B-Thinking",
      "lab": "Baidu",
      "lab_slug": "baidu",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 131072,
      "parameters": "MoE: 21B total, 3B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "ERNIE-4.5-21B-A3B-Thinking is an open-weights deep-thinking model trained from ERNIE-4.5-21B-A3B for reasoning, math, code, and tool use, released in Transformer-style weights for PyTorch and PaddlePaddle tooling.",
      "highlights": [
        "Open-sourced at WAVE SUMMIT 2025 on September 9, 2025",
        "Enhanced 128K long-context understanding",
        "Runs with vLLM, transformers, and FastDeploy"
      ],
      "source_title": "baidu/ERNIE-4.5-21B-A3B-Thinking · Hugging Face",
      "source_url": "https://huggingface.co/baidu/ERNIE-4.5-21B-A3B-Thinking",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#ernie-4-5-21b-a3b-thinking"
    },
    {
      "slug": "seedream-4-0",
      "date": "2025-09-09",
      "model": "Seedream 4.0",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Seedream 4.0 is ByteDance Seed's image creation model that unifies text-to-image generation and general-purpose editing in one architecture, adding commonsense reasoning, multi-image reference, and output up to 4K resolution.",
      "highlights": [
        "Succeeds both Seedream 3.0 and SeedEdit 3.0",
        "Maximum resolution raised from 2K to 4K",
        "Supports in-context reasoning and consistent image sequences",
        "Available on Jimeng, Doubao, and Volcano Ark"
      ],
      "source_title": "Seedream 4.0 Officially Released: Beyond Drawing, Into Imagination",
      "source_url": "https://seed.bytedance.com/en/blog/seedream-4-0-officially-released-beyond-drawing-into-imagination",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#seedream-4-0"
    },
    {
      "slug": "granite-embedding-reranker-r2",
      "date": "2025-09-08",
      "model": "Granite Embedding Reranker R2",
      "lab": "IBM",
      "lab_slug": "ibm",
      "type": "reranker",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "score"
      ],
      "context_window": null,
      "parameters": "149M",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Cross-encoder reranking model from IBM's Granite Embeddings R2 collection that computes relevance scores between queries and documents by encoding them jointly for precise reranking.",
      "highlights": [
        "Part of the Granite Embeddings R2 collection",
        "Cross-encoder architecture",
        "Released under Apache 2.0"
      ],
      "source_title": "ibm-granite/granite-embedding-reranker-english-r2 · Hugging Face",
      "source_url": "https://huggingface.co/ibm-granite/granite-embedding-reranker-english-r2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#granite-embedding-reranker-r2"
    },
    {
      "slug": "hunyuanimage-2-1",
      "date": "2025-09-08",
      "model": "HunyuanImage 2.1",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "17B",
      "model_license": "Tencent Hunyuan Community License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "HunyuanImage 2.1 is an open 17B text-to-image diffusion model that generates 2K (2048x2048) images, using a multimodal LLM and a character-aware multilingual encoder for image-text alignment and text rendering.",
      "highlights": [
        "Native 2K resolution output",
        "FP8 quantized models released September 12, 2025"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/HunyuanImage-2.1: HunyuanImage-2.1: An Efficient Diffusion Model for High-Resolution (2K) Text-to-Image Generation",
      "source_url": "https://github.com/Tencent-Hunyuan/HunyuanImage-2.1",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#hunyuanimage-2-1"
    },
    {
      "slug": "qwen3-asr-flash",
      "date": "2025-09-08",
      "model": "Qwen3-ASR-Flash",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Qwen3-ASR-Flash is a speech recognition service built on Qwen3-Omni and tens of millions of hours of speech data. It supports 11 languages and multiple accents and accepts text context to customize results.",
      "highlights": [
        "Handles singing voice recognition",
        "Users can prompt with context in any format for customized transcription",
        "Offered as an API service"
      ],
      "source_title": "Qwen3 ASR: Hear clearly, transcribe smartly.",
      "source_url": "https://qwen.ai/blog?id=qwen3-asr-flash",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#qwen3-asr-flash"
    },
    {
      "slug": "kimi-k2-instruct-0905",
      "date": "2025-09-05",
      "model": "Kimi K2 Instruct 0905",
      "lab": "Moonshot AI",
      "lab_slug": "moonshot-ai",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "MoE: 1T total, 32B active",
      "model_license": "Modified MIT License",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "Kimi-K2-Instruct-0905 is an updated Kimi K2 instruct model that extends the context window from 128K to 256K tokens. Moonshot also improved its agentic coding and front-end programming abilities.",
      "highlights": [
        "Context length doubled to 256K tokens",
        "Offered on the Kimi Open Platform as kimi-k2-0905-preview",
        "Weights and code under the Modified MIT License"
      ],
      "source_title": "moonshotai/Kimi-K2-Instruct-0905 (Hugging Face)",
      "source_url": "https://huggingface.co/moonshotai/Kimi-K2-Instruct-0905",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#kimi-k2-instruct-0905"
    },
    {
      "slug": "embeddinggemma",
      "date": "2025-09-04",
      "model": "EmbeddingGemma",
      "lab": "Google",
      "lab_slug": "google",
      "type": "embedding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "embedding"
      ],
      "context_window": 2048,
      "parameters": "308M",
      "model_license": "Gemma Terms of Use",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "EmbeddingGemma is a 308M-parameter open text embedding model based on the Gemma 3 architecture and built for on-device use. It is trained on more than 100 languages and runs in under 200MB of RAM when quantized.",
      "highlights": [
        "2K-token context window",
        "768-dimension output, truncatable to 512, 256 or 128 via Matryoshka representation",
        "Weights on Hugging Face, Kaggle and Vertex AI"
      ],
      "source_title": "Introducing EmbeddingGemma: The Best-in-Class Open Model for On-Device Embeddings",
      "source_url": "https://developers.googleblog.com/en/introducing-embeddinggemma/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#embeddinggemma"
    },
    {
      "slug": "hunyuanworld-voyager",
      "date": "2025-09-02",
      "model": "HunyuanWorld-Voyager",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Tencent HunyuanWorld-Voyager Community License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "HunyuanWorld-Voyager is an open video diffusion model that generates world-consistent RGB-D video and 3D point-cloud sequences from a single image along a user-defined camera path, supporting world exploration and 3D reconstruction.",
      "highlights": [
        "Generates aligned depth and RGB video for direct 3D reconstruction",
        "Trained with a data engine that curated over 100,000 video clips"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/HunyuanWorld-Voyager: Voyager is an interactive RGBD video generation model conditioned on camera input, and supports real-time 3D reconstruction.",
      "source_url": "https://github.com/Tencent-Hunyuan/HunyuanWorld-Voyager",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#hunyuanworld-voyager"
    },
    {
      "slug": "hunyuan-mt-7b",
      "date": "2025-09-01",
      "model": "Hunyuan-MT-7B and Hunyuan-MT-Chimera-7B",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "7B",
      "model_license": "Tencent Hunyuan Community License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Hunyuan-MT is Tencent's open translation release, pairing the Hunyuan-MT-7B translation model with Hunyuan-MT-Chimera-7B, an ensemble model that merges multiple translation outputs into a higher-quality result.",
      "highlights": [
        "Supports mutual translation among 33 languages, including five ethnic minority languages in China",
        "Tencent describes Chimera as the industry's first open-source translation ensemble model",
        "FP8 versions also published"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/Hunyuan-MT",
      "source_url": "https://github.com/Tencent-Hunyuan/Hunyuan-MT",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/09#hunyuan-mt-7b"
    },
    {
      "slug": "longcat-flash-chat",
      "date": "2025-08-29",
      "model": "LongCat-Flash-Chat",
      "lab": "Meituan",
      "lab_slug": "meituan",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "MoE: 560B total, 18.6B to 31.3B active",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "LongCat-Flash-Chat is Meituan's first LongCat model, a non-thinking mixture-of-experts foundation model with a dynamic computation mechanism that varies activated parameters per token, aimed at efficient inference and agentic tasks.",
      "highlights": [
        "560B total parameters, activating 18.6B to 31.3B (about 27B on average) per token",
        "Zero-computation experts and a shortcut-connected MoE architecture",
        "128K context at launch; the API platform opened on September 5, 2025",
        "Weights released under MIT"
      ],
      "source_title": "Change Log | LongCat API Docs",
      "source_url": "https://longcat.chat/platform/docs/change-log",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#longcat-flash-chat"
    },
    {
      "slug": "step-audio-2-mini",
      "date": "2025-08-29",
      "model": "Step-Audio 2 mini",
      "lab": "StepFun",
      "lab_slug": "stepfun",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "audio",
        "text"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Step-Audio 2 mini is the open-weight member of StepFun's Step-Audio 2 family, an end-to-end multimodal model for audio understanding and speech conversation. A base version shipped alongside it, and a Think variant followed in September.",
      "highlights": [
        "Step-Audio 2 technical report was published July 23, 2025",
        "Step-Audio 2 mini and Step-Audio 2 mini Base released August 29, 2025",
        "Step-Audio 2 mini Think added September 15, 2025",
        "Released under Apache-2.0"
      ],
      "source_title": "GitHub - stepfun-ai/Step-Audio2",
      "source_url": "https://github.com/stepfun-ai/Step-Audio2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#step-audio-2-mini"
    },
    {
      "slug": "command-a-translate",
      "date": "2025-08-28",
      "model": "Command A Translate",
      "lab": "Cohere",
      "lab_slug": "cohere",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "111B",
      "model_license": "CC-BY-NC-4.0",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "Translation-specialized model in Cohere's Command A family covering 23 business languages, offered with Deep Translation, an agentic multi-step approach that refines translations further.",
      "highlights": [
        "Supports 8K input and 8K output tokens",
        "Deployable on one H100 or A100 with 4-bit quantization",
        "Research weights released as command-a-translate-08-2025"
      ],
      "source_title": "Command A Translate: Secure Enterprise Translation | Cohere",
      "source_url": "https://cohere.com/blog/command-a-translate",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#command-a-translate"
    },
    {
      "slug": "gpt-realtime",
      "date": "2025-08-28",
      "model": "gpt-realtime",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "audio",
        "image"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": 32000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "gpt-realtime is a speech-to-speech model released with the general availability of the Realtime API. OpenAI reports gains in following complex instructions, precise tool calling, and more natural, expressive speech.",
      "highlights": [
        "API model ID gpt-realtime (snapshot gpt-realtime-2025-08-28)",
        "Realtime API added image input, remote MCP servers, and SIP phone calling",
        "Two new voices, Cedar and Marin, exclusive to the Realtime API",
        "32,000-token context window per the API model page"
      ],
      "source_title": "Introducing gpt-realtime and Realtime API updates for production voice agents | OpenAI",
      "source_url": "https://openai.com/index/introducing-gpt-realtime/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#gpt-realtime"
    },
    {
      "slug": "grok-code-fast-1",
      "date": "2025-08-28",
      "model": "Grok Code Fast 1",
      "lab": "xAI",
      "lab_slug": "xai",
      "type": "coding",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Grok Code Fast 1 is a fast, economical reasoning model for agentic coding, built from scratch on a new architecture with a programming-heavy pretraining corpus and post-training on real-world pull requests and coding tasks.",
      "highlights": [
        "API model ID: grok-code-fast-1 (released earlier in stealth as sonic)",
        "Free for a limited time via launch partners including GitHub Copilot and Cursor",
        "Retired May 15, 2026; requests route to grok-build-0.1"
      ],
      "source_title": "Grok Code Fast 1 | SpaceXSI",
      "source_url": "https://x.ai/news/grok-code-fast-1",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#grok-code-fast-1"
    },
    {
      "slug": "hunyuanvideo-foley",
      "date": "2025-08-28",
      "model": "HunyuanVideo-Foley",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "video"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Tencent Hunyuan Community License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "HunyuanVideo-Foley is an open text-video-to-audio model that generates sound effects and audio matched to video content, aimed at short video creation, film production, advertising, and game development.",
      "highlights": [
        "Self-developed 48kHz audio VAE for high-fidelity output",
        "A smaller XL model with offload inference was added September 29, 2025"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/HunyuanVideo-Foley: HunyuanVideo-Foley: Multimodal Diffusion with Representation Alignment for High-Fidelity Foley Audio Generation.",
      "source_url": "https://github.com/Tencent-Hunyuan/HunyuanVideo-Foley",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#hunyuanvideo-foley"
    },
    {
      "slug": "keye-vl-1-5",
      "date": "2025-08-28",
      "model": "Keye-VL-1.5-8B",
      "lab": "Kuaishou",
      "lab_slug": "kuaishou",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "8B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Keye-VL-1.5 is an upgraded 8B multimodal model from Kuaishou's Keye team that improves short-video comprehension with a Slow-Fast video encoding strategy and new long chain-of-thought and reinforcement learning stages.",
      "highlights": [
        "Slow-Fast video encoding strategy for video understanding",
        "Progressive four-stage pre-training",
        "Weights released under Apache-2.0"
      ],
      "source_title": "GitHub - Kwai-Keye/Keye",
      "source_url": "https://github.com/Kwai-Keye/Keye",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#keye-vl-1-5"
    },
    {
      "slug": "mai-1-preview",
      "date": "2025-08-28",
      "model": "MAI-1-preview",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "MAI-1-preview is Microsoft AI's first in-house foundation language model, a mixture-of-experts model released for public testing on LMArena and rolled out for certain text use cases in Copilot.",
      "highlights": [
        "Mixture-of-experts model pre-trained and post-trained on about 15,000 NVIDIA H100 GPUs",
        "Public testing on LMArena, with API access for trusted testers",
        "Rolled out to certain Copilot text use cases"
      ],
      "source_title": "Two in-house models in support of our mission",
      "source_url": "https://microsoft.ai/news/two-new-in-house-models/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#mai-1-preview"
    },
    {
      "slug": "mai-voice-1",
      "date": "2025-08-28",
      "model": "MAI-Voice-1",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "MAI-Voice-1 is Microsoft AI's first in-house speech generation model, producing expressive audio for single and multi-speaker scenarios. It launched in Copilot Daily and Podcasts and later reached Microsoft Foundry.",
      "highlights": [
        "Generates a minute of audio in under a second on a single GPU, per Microsoft",
        "Launched in Copilot Daily, Copilot Podcasts and Copilot Labs",
        "Added to Microsoft Foundry with custom voice creation on April 2, 2026"
      ],
      "source_title": "Two in-house models in support of our mission",
      "source_url": "https://microsoft.ai/news/two-new-in-house-models/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#mai-voice-1"
    },
    {
      "slug": "olmoasr",
      "date": "2025-08-28",
      "model": "OLMoASR",
      "lab": "Ai2",
      "lab_slug": "ai2",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "244M to 1.5B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Family of fully open English automatic speech recognition models from Ai2, trained from scratch on a curated, large-scale dataset and released in sizes from 244M to 1.5B.",
      "highlights": [
        "Sizes include small.en at 244M, medium.en at 769M and two 1.5B large.en models",
        "The large.en-v2 model was trained on 680,000 hours of audio",
        "Released under Apache 2.0"
      ],
      "source_title": "OLMoASR: A series of open speech recognition models | Ai2",
      "source_url": "https://allenai.org/blog/olmoasr",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#olmoasr"
    },
    {
      "slug": "gemini-2-5-flash-image",
      "date": "2025-08-26",
      "model": "Gemini 2.5 Flash Image (Nano Banana)",
      "lab": "Google",
      "lab_slug": "google",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text",
        "image"
      ],
      "context_window": 65536,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms"
      ],
      "summary": "Gemini 2.5 Flash Image, nicknamed Nano Banana, is Google's image generation and editing model that can blend multiple images, keep characters consistent and make targeted edits with natural language.",
      "highlights": [
        "Preview model gemini-2.5-flash-image-preview in the Gemini API, Google AI Studio and Vertex AI",
        "Generally available on Oct 2, 2025",
        "All generated or edited images carry an invisible SynthID watermark"
      ],
      "source_title": "Introducing Gemini 2.5 Flash Image, our state-of-the-art image model",
      "source_url": "https://developers.googleblog.com/introducing-gemini-2-5-flash-image/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#gemini-2-5-flash-image"
    },
    {
      "slug": "hermes-4",
      "date": "2025-08-26",
      "model": "Hermes 4",
      "lab": "Nous Research",
      "lab_slug": "nous-research",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "14B, 70B and 405B",
      "model_license": "Llama 3 Community License (70B, 405B); Apache-2.0 (14B)",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Hybrid-mode reasoning models from Nous Research released with a technical report, led by a frontier model based on Llama 3.1 405B.",
      "highlights": [
        "Released as Hermes-4-Llama-3.1-405B, Hermes-4-Llama-3.1-70B and Hermes-4-14B",
        "14B model is a small dense variant for local inference",
        "Hermes 4 Technical Report published the same day"
      ],
      "source_title": "Releases | Nous Research",
      "source_url": "https://nousresearch.com/releases",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#hermes-4"
    },
    {
      "slug": "wan2-2-s2v",
      "date": "2025-08-26",
      "model": "Wan2.2-S2V-14B",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "audio"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": "14B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Wan2.2-S2V-14B is an audio-driven cinematic video generation model in the Wan2.2 family that generates video from speech and a reference image.",
      "highlights": [
        "Supports 480P and 720P output",
        "Text-to-speech support through CosyVoice added September 5, 2025",
        "Released under the Apache 2.0 license"
      ],
      "source_title": "Wan2.2 (GitHub)",
      "source_url": "https://github.com/Wan-Video/Wan2.2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#wan2-2-s2v"
    },
    {
      "slug": "grok-2-open-weights",
      "date": "2025-08-23",
      "model": "Grok 2 (open weights)",
      "lab": "xAI",
      "lab_slug": "xai",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Grok 2 Community License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "xAI published the weights of Grok 2, a post-trained model it trained and used in 2024, on Hugging Face, with instructions for serving the roughly 500 GB checkpoint with SGLang across eight GPUs.",
      "highlights": [
        "Hugging Face repo: xai-org/grok-2 (42 files, about 500 GB)",
        "Config lists a mixture of 8 experts with 2 active per token",
        "Checkpoint is TP=8 and needs 8 GPUs with more than 40 GB memory each",
        "LICENSE file was retitled xAI Community License Agreement on November 4, 2025"
      ],
      "source_title": "xai-org/grok-2 · Hugging Face",
      "source_url": "https://huggingface.co/xai-org/grok-2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#grok-2-open-weights"
    },
    {
      "slug": "command-a-reasoning",
      "date": "2025-08-21",
      "model": "Command A Reasoning",
      "lab": "Cohere",
      "lab_slug": "cohere",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "111B",
      "model_license": "CC-BY-NC-4.0",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "Cohere's first reasoning model, built for enterprise agentic workflows with a user-set token budget that switches between maximum accuracy and higher throughput.",
      "highlights": [
        "Runs on a single H100 or A100 at 128K context",
        "Core generative model behind the North agent platform",
        "Research weights released as command-a-reasoning-08-2025"
      ],
      "source_title": "Command A Reasoning: Control for AI Agents | Cohere",
      "source_url": "https://cohere.com/blog/command-a-reasoning",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#command-a-reasoning"
    },
    {
      "slug": "deepseek-v3-1",
      "date": "2025-08-21",
      "model": "DeepSeek-V3.1",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "MoE: 671B total, 37B active",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "DeepSeek-V3.1 is a hybrid model that supports thinking and non-thinking modes in one model, with stronger tool use and multi-step agent skills from post-training. DeepSeek called it its first step toward the agent era.",
      "highlights": [
        "V3.1 Base adds 840B tokens of continued pre-training for long-context extension on top of V3",
        "128K context for both deepseek-chat and deepseek-reasoner",
        "API added Anthropic format support and strict function calling in beta"
      ],
      "source_title": "DeepSeek-V3.1 Release",
      "source_url": "https://api-docs.deepseek.com/news/news250821",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#deepseek-v3-1"
    },
    {
      "slug": "seed-oss-36b",
      "date": "2025-08-20",
      "model": "Seed-OSS-36B",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 512000,
      "parameters": "36B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Seed-OSS-36B is an open-weights language model series from ByteDance Seed, trained on 12T tokens with native 512K context, a user-controllable thinking budget, and tuning for reasoning and agent tasks.",
      "highlights": [
        "Released as Base, Base-woSyn (no synthetic instruction data), and Instruct",
        "Native 512K-token context",
        "Thinking budget can be adjusted flexibly at inference",
        "Weights on Hugging Face under Apache-2.0"
      ],
      "source_title": "ByteDance-Seed/Seed-OSS-36B-Instruct · Hugging Face",
      "source_url": "https://huggingface.co/ByteDance-Seed/Seed-OSS-36B-Instruct",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#seed-oss-36b"
    },
    {
      "slug": "qwen-image-edit",
      "date": "2025-08-19",
      "model": "Qwen-Image-Edit",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "20B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Consumer app",
        "Open weights"
      ],
      "summary": "Qwen-Image-Edit is an image editing model built on the 20B Qwen-Image model, extending its text rendering to precise text editing in images. It feeds the input image to Qwen2.5-VL and a VAE encoder for both semantic and appearance editing.",
      "highlights": [
        "Available through the Image Editing feature on Qwen Chat",
        "Weights on Hugging Face and ModelScope",
        "Released under the Apache 2.0 license"
      ],
      "source_title": "Qwen-Image-Edit: Image Editing with Higher Quality and Efficiency",
      "source_url": "https://qwenlm.github.io/blog/qwen-image-edit/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#qwen-image-edit"
    },
    {
      "slug": "nemotron-nano-2",
      "date": "2025-08-18",
      "model": "NVIDIA Nemotron Nano 2",
      "lab": "NVIDIA",
      "lab_slug": "nvidia",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "9B",
      "model_license": "NVIDIA Open Model License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "9B language model trained from scratch by NVIDIA as a unified model for reasoning and non-reasoning tasks, generating a reasoning trace before its final answer.",
      "highlights": [
        "Supports English, German, Spanish, French, Italian and Japanese",
        "A 12B v2 variant was released August 29, 2025",
        "Released under the NVIDIA Open Model License"
      ],
      "source_title": "nvidia/NVIDIA-Nemotron-Nano-9B-v2 · Hugging Face",
      "source_url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-Nano-9B-v2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#nemotron-nano-2"
    },
    {
      "slug": "granite-embedding-r2",
      "date": "2025-08-15",
      "model": "Granite Embedding R2",
      "lab": "IBM",
      "lab_slug": "ibm",
      "type": "embedding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "embedding"
      ],
      "context_window": 8192,
      "parameters": "149M and 47M",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Second-generation English Granite embedding models based on ModernBERT, with a 149M model replacing the 125M version and a smaller 47M model.",
      "highlights": [
        "8,192-token context length",
        "149M model outputs 768-dimensional vectors; 47M outputs 384",
        "Released under Apache 2.0"
      ],
      "source_title": "ibm-granite/granite-embedding-english-r2 · Hugging Face",
      "source_url": "https://huggingface.co/ibm-granite/granite-embedding-english-r2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#granite-embedding-r2"
    },
    {
      "slug": "gemma-3-270m",
      "date": "2025-08-14",
      "model": "Gemma 3 270M",
      "lab": "Google",
      "lab_slug": "google",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32768,
      "parameters": "270M",
      "model_license": "Gemma Terms of Use",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "Gemma 3 270M is a compact Gemma 3 model designed as a base for task-specific fine-tuning, such as classification and entity extraction. Google calls it the most power-efficient Gemma model to date.",
      "highlights": [
        "170M embedding parameters and 100M transformer parameters, with a 256K-token vocabulary",
        "Downloads on Hugging Face, Ollama, Kaggle, LM Studio and Docker",
        "Pretrained, instruction-tuned and QAT checkpoints released"
      ],
      "source_title": "Introducing Gemma 3 270M: The compact model for hyper-efficient AI",
      "source_url": "https://developers.googleblog.com/en/introducing-gemma-3-270m/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#gemma-3-270m"
    },
    {
      "slug": "hunyuan-gamecraft-1-0",
      "date": "2025-08-14",
      "model": "Hunyuan-GameCraft-1.0",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Tencent Hunyuan Community License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Hunyuan-GameCraft is an open framework for high-dynamic interactive game video generation that maps keyboard and mouse inputs into a shared camera representation for fine-grained action control and long video extension.",
      "highlights": [
        "Generates from a reference image, a text prompt, and keyboard or mouse signals",
        "Gradio demo code followed on August 21, 2025"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/Hunyuan-GameCraft-1.0: Hunyuan-GameCraft: High-dynamic Interactive Game Video Generation with Hybrid History Condition",
      "source_url": "https://github.com/Tencent-Hunyuan/Hunyuan-GameCraft-1.0",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#hunyuan-gamecraft-1-0"
    },
    {
      "slug": "nextstep-1",
      "date": "2025-08-14",
      "model": "NextStep-1",
      "lab": "StepFun",
      "lab_slug": "stepfun",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "14B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "NextStep-1 is a 14B-parameter autoregressive image generation model that works directly with continuous image tokens rather than discrete ones, released with text-to-image and image editing checkpoints.",
      "highlights": [
        "Released as NextStep-1-Large-Pretrain, NextStep-1-Large, and NextStep-1-Large-Edit",
        "Autoregressive generation over continuous image tokens",
        "Weights on Hugging Face under Apache-2.0"
      ],
      "source_title": "GitHub - stepfun-ai/NextStep-1",
      "source_url": "https://github.com/stepfun-ai/NextStep-1",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#nextstep-1"
    },
    {
      "slug": "parakeet-v3-canary-v2",
      "date": "2025-08-14",
      "model": "Parakeet TDT 0.6B v3 and Canary-1B v2",
      "lab": "NVIDIA",
      "lab_slug": "nvidia",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "600M and 1B",
      "model_license": "CC-BY-4.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Multilingual speech models covering 25 European languages: Parakeet TDT 0.6B v3 for high-throughput transcription and Canary-1B v2 for transcription and translation.",
      "highlights": [
        "Canary-1B v2 translates between English and 24 languages",
        "Parakeet v3 extends the English-only v2",
        "Both released under CC BY 4.0"
      ],
      "source_title": "nvidia/canary-1b-v2 · Hugging Face",
      "source_url": "https://huggingface.co/nvidia/canary-1b-v2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#parakeet-v3-canary-v2"
    },
    {
      "slug": "lfm2-vl",
      "date": "2025-08-12",
      "model": "LFM2-VL (450M, 1.6B)",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32768,
      "parameters": "450M; 1.6B",
      "model_license": "LFM Open License v1.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Liquid AI's first vision-language models, extending LFM2 with SigLIP2 NaFlex image encoders to process text and images at native resolution, released as open-weight 450M and 1.6B variants for on-device use.",
      "highlights": [
        "Native image resolution up to 512x512; larger images split into 512x512 patches",
        "Built on LFM2-350M (450M model) and LFM2-1.2B (1.6B model) backbones",
        "User-tunable image token count trades speed for quality at inference",
        "Weights on Hugging Face; supported by LEAP for edge deployment"
      ],
      "source_title": "LFM2-VL: Efficient Vision-Language Models",
      "source_url": "https://www.liquid.ai/blog/lfm2-vl-efficient-vision-language-models",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#lfm2-vl"
    },
    {
      "slug": "mistral-medium-3-1",
      "date": "2025-08-12",
      "model": "Mistral Medium 3.1",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "August 2025 update of Mistral Medium 3, Mistral's frontier-class multimodal model, with improvements to tone and performance and the same 128K-token context window.",
      "highlights": [
        "API model ID mistral-medium-2508",
        "128K-token context window",
        "Deprecated May 22, 2026 in favor of Mistral Medium 3.5"
      ],
      "source_title": "Mistral Medium 3.1 | Mistral Docs",
      "source_url": "https://docs.mistral.ai/models/mistral-medium-3-1-25-08",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#mistral-medium-3-1"
    },
    {
      "slug": "molmoact",
      "date": "2025-08-12",
      "model": "MolmoAct",
      "lab": "Ai2",
      "lab_slug": "ai2",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "action"
      ],
      "context_window": null,
      "parameters": "7B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Ai2's first Action Reasoning Model, a robotics model that reasons about its environment in 3D before acting, released with an open post-training dataset.",
      "highlights": [
        "Post-trained on about 10,000 distinct robot episodes",
        "Pretrained on a curated subset of Open-X Embodiment data",
        "Released under Apache 2.0"
      ],
      "source_title": "MolmoAct: An Action Reasoning Model that reasons in 3D space | Ai2",
      "source_url": "https://allenai.org/blog/molmoact",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#molmoact"
    },
    {
      "slug": "glm-4-5v",
      "date": "2025-08-11",
      "model": "GLM-4.5V",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video",
        "file"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "MoE: 106B total, 12B active",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "GLM-4.5V is Z.ai's open-weights vision-language reasoning model built on the GLM-4.5-Air text model. It covers image, video, and document understanding, GUI agent tasks, and front-end replication from images.",
      "highlights": [
        "Adds a thinking mode switch for fast responses or deep reasoning",
        "Continues the technical approach of GLM-4.1V-Thinking",
        "Weights under the MIT license"
      ],
      "source_title": "GLM-V (GitHub)",
      "source_url": "https://github.com/zai-org/GLM-V",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#glm-4-5v"
    },
    {
      "slug": "gpt-5",
      "date": "2025-08-07",
      "model": "GPT-5 (with GPT-5 mini and GPT-5 nano)",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 400000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API",
        "Enterprise"
      ],
      "summary": "GPT-5 is OpenAI's flagship model family, launched in ChatGPT as a unified system that routes between a fast model and GPT-5 thinking, and in the API as gpt-5, gpt-5-mini, and gpt-5-nano with configurable reasoning effort.",
      "highlights": [
        "API models gpt-5, gpt-5-mini, and gpt-5-nano; ChatGPT's non-reasoning model is gpt-5-chat-latest",
        "API context totals 400,000 tokens (272,000 input plus 128,000 reasoning and output)",
        "GPT-5 pro for ChatGPT Pro users; gpt-5-pro reached the API on October 6, 2025",
        "API added a minimal reasoning effort and custom tool calls"
      ],
      "source_title": "Introducing GPT-5 | OpenAI",
      "source_url": "https://openai.com/index/introducing-gpt-5/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#gpt-5"
    },
    {
      "slug": "minimax-speech-2-5",
      "date": "2025-08-07",
      "model": "MiniMax Speech 2.5",
      "lab": "MiniMax",
      "lab_slug": "minimax",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "MiniMax Speech 2.5 is a text-to-speech model that builds on Speech 02 with stronger multilingual expressiveness, more accurate voice cloning, and language support expanded to more than 40 languages.",
      "highlights": [
        "Voice cloning preserves accents and speech habits of the source voice",
        "Expanded language support to over 40 languages"
      ],
      "source_title": "MiniMax Speech 2.5 Launches: Enhanced Multilingual Expressiveness Exceptional Voice Cloning Fidelity",
      "source_url": "https://www.minimax.io/news/minimax-speech-25",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#minimax-speech-2-5"
    },
    {
      "slug": "claude-opus-4-1",
      "date": "2025-08-05",
      "model": "Claude Opus 4.1",
      "lab": "Anthropic",
      "lab_slug": "anthropic",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Cloud platforms"
      ],
      "summary": "Claude Opus 4.1 is an incremental upgrade to Claude Opus 4 with improvements in agentic tasks, real-world coding, and reasoning, including better detail tracking and agentic search for research and data analysis.",
      "highlights": [
        "API model ID claude-opus-4-1-20250805",
        "Available to paid Claude users, in Claude Code, and on the Claude API, Amazon Bedrock, and Vertex AI",
        "Retired on the Claude API on August 5, 2026"
      ],
      "source_title": "Claude Opus 4.1",
      "source_url": "https://www.anthropic.com/news/claude-opus-4-1",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#claude-opus-4-1"
    },
    {
      "slug": "eleven-music",
      "date": "2025-08-05",
      "model": "Eleven Music",
      "lab": "ElevenLabs",
      "lab_slug": "elevenlabs",
      "type": "music",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "ElevenLabs' first music model, generating studio-grade songs from natural language prompts with control over genre, style and structure, with or without vocals, in multiple languages.",
      "highlights": [
        "Section-level editing of sound and lyrics",
        "Available in the API from August 18, 2025",
        "Model ID music_v1, since superseded by Music v2"
      ],
      "source_title": "Eleven Music is here",
      "source_url": "https://elevenlabs.io/blog/eleven-music-is-here",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#eleven-music"
    },
    {
      "slug": "gpt-oss",
      "date": "2025-08-05",
      "model": "gpt-oss-120b and gpt-oss-20b",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 131072,
      "parameters": "120b: 117B total, 5.1B active; 20b: 21B total, 3.6B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "gpt-oss-120b and gpt-oss-20b are OpenAI's open-weight mixture-of-experts reasoning models, released under Apache 2.0 with adjustable reasoning effort, full chain-of-thought, tool use, and Structured Outputs support.",
      "highlights": [
        "Natively quantized in MXFP4: 120b runs within 80 GB of memory, 20b within 16 GB",
        "Weights on Hugging Face; launch partners included Azure, AWS, vLLM, and Ollama",
        "Trained on the harmony response format; o200k_harmony tokenizer open-sourced",
        "Native 128k context; API docs list a 131,072-token window"
      ],
      "source_title": "Introducing gpt-oss | OpenAI",
      "source_url": "https://openai.com/index/introducing-gpt-oss/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#gpt-oss"
    },
    {
      "slug": "leonardo-lucid-origin",
      "date": "2025-08-05",
      "model": "Lucid Origin",
      "lab": "Leonardo.Ai",
      "lab_slug": "leonardo-ai",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Leonardo.Ai foundation image model trained for Full HD renders with greater diversity and dynamic range, balancing strong prompt adherence and text rendering with stylistic range.",
      "highlights": [
        "Trained for Full HD renders",
        "Built by blending training checkpoints in a curated process",
        "Offered in Fast and Ultra modes on Leonardo"
      ],
      "source_title": "Lucid Origin - Leonardo.AI's latest Gen AI model",
      "source_url": "https://leonardo.ai/news/lucid-origin-ai-image-model",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#leonardo-lucid-origin"
    },
    {
      "slug": "hunyuan-dense-0-5b-to-7b",
      "date": "2025-08-04",
      "model": "Hunyuan 0.5B, 1.8B, 4B, and 7B",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "0.5B, 1.8B, 4B, 7B",
      "model_license": "Tencent Hunyuan Community License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Tencent open-sourced four dense Hunyuan language models at 0.5B, 1.8B, 4B, and 7B parameters, each with pretrained and instruct versions, trained with strategies similar to Hunyuan-A13B for deployment from edge devices to servers.",
      "highlights": [
        "Hybrid fast and slow thinking modes",
        "Native 256K context window",
        "FP8, GPTQ-Int4, and AWQ-Int4 quantized versions published"
      ],
      "source_title": "LICENSE · tencent/Hunyuan-7B-Instruct at main",
      "source_url": "https://huggingface.co/tencent/Hunyuan-7B-Instruct/blob/main/LICENSE",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#hunyuan-dense-0-5b-to-7b"
    },
    {
      "slug": "qwen-image",
      "date": "2025-08-04",
      "model": "Qwen-Image",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "20B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Consumer app",
        "Open weights"
      ],
      "summary": "Qwen-Image is a 20B MMDiT image foundation model with significant advances in complex text rendering and precise image editing. It renders both alphabetic and logographic text, including English and Chinese.",
      "highlights": [
        "Weights on Hugging Face and ModelScope at launch",
        "Image Generation available on Qwen Chat",
        "Released under the Apache 2.0 license"
      ],
      "source_title": "Qwen-Image: Crafting with Native Text Rendering",
      "source_url": "https://qwenlm.github.io/blog/qwen-image/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#qwen-image"
    },
    {
      "slug": "gemini-2-5-deep-think",
      "date": "2025-08-01",
      "model": "Gemini 2.5 Deep Think",
      "lab": "Google",
      "lab_slug": "google",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Gemini 2.5 Deep Think is an enhanced reasoning mode for Gemini 2.5 that uses parallel thinking to explore several ideas at once before answering, rolled out in the Gemini app to Google AI Ultra subscribers.",
      "highlights": [
        "First previewed at Google I/O on May 20, 2025 as an experimental mode for trusted testers",
        "Turned on with a Deep Think toggle in the Gemini app prompt bar",
        "A research version was shared with a small group of mathematicians and academics"
      ],
      "source_title": "Gemini 2.5: Deep Think is now rolling out",
      "source_url": "https://blog.google/products-and-platforms/products/gemini/gemini-2-5-deep-think/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#gemini-2-5-deep-think"
    },
    {
      "slug": "granite-guardian-3-3",
      "date": "2025-08-01",
      "model": "Granite Guardian 3.3",
      "lab": "IBM",
      "lab_slug": "ibm",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "8B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Granite 3.3 8B-based safety model that judges whether prompts and responses meet specified criteria, with built-in checks for jailbreaks, harm and RAG groundedness.",
      "highlights": [
        "Based on Granite 3.3 8B",
        "Supports custom criteria beyond pre-baked detectors",
        "Released under Apache 2.0"
      ],
      "source_title": "ibm-granite/granite-guardian-3.3-8b · Hugging Face",
      "source_url": "https://huggingface.co/ibm-granite/granite-guardian-3.3-8b",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/08#granite-guardian-3-3"
    },
    {
      "slug": "command-a-vision",
      "date": "2025-07-31",
      "model": "Command A Vision",
      "lab": "Cohere",
      "lab_slug": "cohere",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "112B",
      "model_license": "CC-BY-NC-4.0",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "Cohere's first Command model that understands images, built for enterprise visual tasks such as charts, diagrams, PDFs, slides and real-world scenes.",
      "highlights": [
        "Deploys privately on two A100 GPUs or one H100 with 4-bit quantization",
        "Keeps Command A's RAG with citations",
        "Research weights released as command-a-vision-07-2025"
      ],
      "source_title": "Command A Vision: Multimodal AI for Business | Cohere",
      "source_url": "https://cohere.com/blog/command-a-vision",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#command-a-vision"
    },
    {
      "slug": "flux-1-krea-dev",
      "date": "2025-07-31",
      "model": "FLUX.1 Krea [dev]",
      "lab": "Black Forest Labs",
      "lab_slug": "black-forest-labs",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "FLUX.1 [dev] Non-Commercial License",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "Open-weight text-to-image model developed with Krea AI as the open version of Krea 1, trained to avoid the oversaturated AI look and produce more realistic, diverse images.",
      "highlights": [
        "Architecturally compatible with the FLUX.1 [dev] ecosystem",
        "Weights on Hugging Face, commercial licenses via the BFL Licensing Portal",
        "Positioned as a base model for downstream fine-tuning"
      ],
      "source_title": "FLUX.1 Krea [dev]: An 'Opinionated' Text-to-Image Model",
      "source_url": "https://bfl.ai/blog/flux-1-krea-dev",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#flux-1-krea-dev"
    },
    {
      "slug": "codestral-2508",
      "date": "2025-07-30",
      "model": "Codestral 25.08",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "coding",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Enterprise"
      ],
      "summary": "Update to the Codestral fill-in-the-middle model, announced as part of Mistral's coding stack for enterprises with more accepted completions and fewer runaway generations.",
      "highlights": [
        "Deployable in cloud, VPC or on-premises environments",
        "Improved chat-mode instruction following",
        "API model ID codestral-2508"
      ],
      "source_title": "Announcing Codestral 25.08 and the Complete Mistral Coding Stack for Enterprise | Mistral AI",
      "source_url": "https://mistral.ai/news/codestral-25-08",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#codestral-2508"
    },
    {
      "slug": "glm-4-5",
      "date": "2025-07-28",
      "model": "GLM-4.5",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "MoE: 355B total, 32B active; Air 106B total, 12B active",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "GLM-4.5 is Z.ai's open-weights foundation model series for agents, unifying reasoning, coding, and agentic abilities. It comes in GLM-4.5 and the more compact GLM-4.5-Air, both with thinking and non-thinking modes.",
      "highlights": [
        "Pre-trained on 15T tokens of general data, then on code, reasoning, and agent data",
        "128K context length",
        "Weights on Hugging Face and ModelScope under the MIT license"
      ],
      "source_title": "GLM-4.5: Reasoning, Coding, and Agentic Abilities",
      "source_url": "https://z.ai/blog/glm-4.5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#glm-4-5"
    },
    {
      "slug": "wan2-2",
      "date": "2025-07-28",
      "model": "Wan2.2",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": "MoE: A14B (T2V, I2V); 5B (TI2V)",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Wan2.2 is the open video generation family from Alibaba's Wan team, adding mixture-of-experts text-to-video and image-to-video models and a compact 5B text-and-image-to-video model.",
      "highlights": [
        "T2V-A14B and I2V-A14B support 480P and 720P",
        "TI2V-5B uses a high-compression VAE and supports 720P",
        "Integrated into Diffusers and ComfyUI at launch; Apache 2.0 license"
      ],
      "source_title": "Wan2.2 (GitHub)",
      "source_url": "https://github.com/Wan-Video/Wan2.2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#wan2-2"
    },
    {
      "slug": "hunyuanworld-1-0",
      "date": "2025-07-26",
      "model": "HunyuanWorld 1.0",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "3d"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Tencent HunyuanWorld-1.0 Community License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "HunyuanWorld 1.0 is Tencent's open model for generating immersive, explorable 360-degree 3D worlds from text or images, combining panoramic proxy generation, semantic layering, and hierarchical 3D reconstruction.",
      "highlights": [
        "Tencent describes it as the first open-source, simulation-capable immersive 3D world generation model",
        "Quantized HunyuanWorld-1.0-lite for consumer GPUs added August 15, 2025"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/HunyuanWorld-1.0: Generating Immersive, Explorable, and Interactive 3D Worlds from Words or Pixels with Hunyuan3D World Model",
      "source_url": "https://github.com/Tencent-Hunyuan/HunyuanWorld-1.0",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#hunyuanworld-1-0"
    },
    {
      "slug": "runway-aleph",
      "date": "2025-07-25",
      "model": "Aleph",
      "lab": "Runway",
      "lab_slug": "runway",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "In-context video model for multi-task editing that adds, removes and transforms objects, generates new angles of a scene and modifies style and lighting on an input video.",
      "highlights": [
        "Available to all paid plans from July 31, 2025",
        "Generates new camera angles from existing footage",
        "Succeeded by Aleph 2.0 in May 2026"
      ],
      "source_title": "Introducing Runway Aleph",
      "source_url": "https://runway.com/research/introducing-runway-aleph",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#runway-aleph"
    },
    {
      "slug": "llama-3-3-nemotron-super-49b-v1-5",
      "date": "2025-07-25",
      "model": "Llama-3.3-Nemotron-Super-49B-v1.5",
      "lab": "NVIDIA",
      "lab_slug": "nvidia",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "49B",
      "model_license": "NVIDIA Open Model License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Significantly upgraded version of Llama-3.3-Nemotron-Super-49B, a reasoning model NVIDIA derived from Meta's Llama-3.3-70B-Instruct using neural architecture search to cut memory needs while keeping accuracy.",
      "highlights": [
        "Trained between November 2024 and July 2025",
        "Post-trained for reasoning, chat preferences and agentic tasks such as RAG and tool calling",
        "Part of the Llama Nemotron collection"
      ],
      "source_title": "nvidia/Llama-3_3-Nemotron-Super-49B-v1_5 · Hugging Face",
      "source_url": "https://huggingface.co/nvidia/Llama-3_3-Nemotron-Super-49B-v1_5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#llama-3-3-nemotron-super-49b-v1-5"
    },
    {
      "slug": "qwen3-thinking-2507",
      "date": "2025-07-25",
      "model": "Qwen3-235B-A22B-Thinking-2507",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 262144,
      "parameters": "MoE: 235B total, 22B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Qwen3-235B-A22B-Thinking-2507 is an updated thinking-mode version of Qwen3-235B-A22B, released as a separate reasoning model in the Qwen3-2507 series, which split the hybrid Qwen3 model into dedicated instruct and thinking models.",
      "highlights": [
        "Native 262,144-token context",
        "Qwen3-30B-A3B-Thinking-2507 followed on July 31 and Qwen3-4B-Thinking-2507 on August 6, 2025",
        "Weights under the Apache 2.0 license"
      ],
      "source_title": "Qwen/Qwen3-235B-A22B-Thinking-2507 (Hugging Face)",
      "source_url": "https://huggingface.co/Qwen/Qwen3-235B-A22B-Thinking-2507",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#qwen3-thinking-2507"
    },
    {
      "slug": "step-3",
      "date": "2025-07-25",
      "model": "Step 3",
      "lab": "StepFun",
      "lab_slug": "stepfun",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 65536,
      "parameters": "MoE: 321B total, 38B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Step 3 is StepFun's multimodal reasoning model built on a mixture-of-experts architecture and co-designed with Multi-Matrix Factorization Attention and Attention-FFN Disaggregation to reduce decoding costs.",
      "highlights": [
        "321B total parameters (316B language model) with 38B active per token",
        "Maximum context length of 65,536 tokens",
        "Model weights were uploaded to Hugging Face on July 31, 2025 under Apache-2.0"
      ],
      "source_title": "stepfun-ai/step3 · Hugging Face",
      "source_url": "https://huggingface.co/stepfun-ai/step3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#step-3"
    },
    {
      "slug": "qwen-mt",
      "date": "2025-07-24",
      "model": "Qwen-MT",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Qwen-MT (qwen-mt-turbo) is Qwen's translation model on the Qwen API, built on Qwen3 and trained on trillions of multilingual and translation tokens. It uses a lightweight mixture-of-experts architecture for low latency.",
      "highlights": [
        "Translates across 92 major languages and dialects",
        "Supports terminology intervention, domain prompts, and translation memory",
        "Available through Alibaba Cloud Model Studio"
      ],
      "source_title": "Qwen-MT: Where Speed Meets Smart Translation",
      "source_url": "https://qwenlm.github.io/blog/qwen-mt/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#qwen-mt"
    },
    {
      "slug": "seed-liveinterpret-2-0",
      "date": "2025-07-24",
      "model": "Seed LiveInterpret 2.0",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "audio",
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Seed LiveInterpret 2.0 is an end-to-end simultaneous interpretation model for Chinese-to-English and English-to-Chinese speech, built on a full-duplex framework with real-time voice cloning so translations keep the speaker's voice.",
      "highlights": [
        "Covers Chinese-to-English and English-to-Chinese interpretation",
        "Clones the speaker's voice in real time",
        "Offered on Volcano Engine as Doubao Simultaneous Interpretation 2.0",
        "Ola Friend earbuds announced as the first hardware to adopt it"
      ],
      "source_title": "Seed LiveInterpret 2.0 Released: An End-to-End Simultaneous Interpretation Model Featuring Ultra-High Accuracy Close to Human Interpreters, Low Latency of 3 Seconds, and Real-Time Voice Cloning",
      "source_url": "https://seed.bytedance.com/en/blog/seed-liveinterpret-2-0-released-an-end-to-end-simultaneous-interpretation-model-featuring-ultra-high-accuracy-close-to-human-interpreters-low-latency-of-3-seconds-and-real-time-voice-cloning",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#seed-liveinterpret-2-0"
    },
    {
      "slug": "qwen3-coder",
      "date": "2025-07-22",
      "model": "Qwen3-Coder",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 262144,
      "parameters": "MoE: 480B total, 35B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "Qwen3-Coder is Alibaba's agentic coding model family, led by Qwen3-Coder-480B-A35B-Instruct. Alibaba also open-sourced Qwen Code, a command-line tool for agentic coding in the terminal, alongside it.",
      "highlights": [
        "Pre-trained on 7.5T tokens with a 70% code ratio",
        "Native 256K context, extendable to 1M with YaRN",
        "Qwen3-Coder-30B-A3B-Instruct followed on July 31, 2025",
        "Weights under the Apache 2.0 license"
      ],
      "source_title": "Qwen3-Coder: Agentic Coding in the World",
      "source_url": "https://qwenlm.github.io/blog/qwen3-coder/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#qwen3-coder"
    },
    {
      "slug": "qwen3-instruct-2507",
      "date": "2025-07-21",
      "model": "Qwen3-235B-A22B-Instruct-2507",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 262144,
      "parameters": "MoE: 235B total, 22B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Qwen3-235B-A22B-Instruct-2507 is an updated non-thinking version of Qwen3-235B-A22B with significant improvements and 256K-token long-context understanding. It began the Qwen3-2507 series of separate instruct and thinking models.",
      "highlights": [
        "Native 262,144-token context; 1M-token support added August 8, 2025",
        "Qwen3-30B-A3B-Instruct-2507 followed on July 30 and Qwen3-4B-Instruct-2507 on August 6, 2025",
        "Weights under the Apache 2.0 license"
      ],
      "source_title": "Qwen3 (GitHub)",
      "source_url": "https://github.com/QwenLM/Qwen3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#qwen3-instruct-2507"
    },
    {
      "slug": "seed-x",
      "date": "2025-07-18",
      "model": "Seed-X",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "7B",
      "model_license": "OpenMDW-1.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Seed-X is an open-weights family of 7B multilingual translation models from ByteDance Seed, comprising instruction, reinforcement-learning, and reward models, supporting translation among 28 languages.",
      "highlights": [
        "Seed-X-Instruct, Seed-X-PPO, and Seed-X-RM released on Hugging Face",
        "Uses a Mistral-style architecture",
        "Quantized 8-bit and 4-bit PPO models followed on July 28, 2025"
      ],
      "source_title": "GitHub - ByteDance-Seed/Seed-X-7B",
      "source_url": "https://github.com/ByteDance-Seed/Seed-X-7B",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#seed-x"
    },
    {
      "slug": "runway-act-two",
      "date": "2025-07-17",
      "model": "Act-Two",
      "lab": "Runway",
      "lab_slug": "runway",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "video",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Next-generation motion capture model succeeding Act-One, with major gains in generation quality and support for head, face, body and hand tracking for character performances.",
      "highlights": [
        "Head, face, body and hand tracking",
        "Available on paid plans",
        "Voice changing added to the Act-Two interface in August 2025"
      ],
      "source_title": "Product Updates & Changelog | Runway AI",
      "source_url": "https://runway.com/changelog",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#runway-act-two"
    },
    {
      "slug": "canary-qwen-2-5b",
      "date": "2025-07-17",
      "model": "Canary-Qwen-2.5B",
      "lab": "NVIDIA",
      "lab_slug": "nvidia",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "2.5B",
      "model_license": "CC-BY-4.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "English speech recognition model built as a speech-augmented language model from a FastConformer Canary encoder and the Qwen3-1.7B language model.",
      "highlights": [
        "Combines canary-1b-flash with Qwen3-1.7B through LoRA",
        "Supports English speech-to-text",
        "Released under CC BY 4.0"
      ],
      "source_title": "nvidia/canary-qwen-2.5b · Hugging Face",
      "source_url": "https://huggingface.co/nvidia/canary-qwen-2.5b",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#canary-qwen-2-5b"
    },
    {
      "slug": "ltx-video-0-9-8",
      "date": "2025-07-16",
      "model": "LTX-Video 0.9.8",
      "lab": "Lightricks",
      "lab_slug": "lightricks",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": "13B and 2B",
      "model_license": "LTXV Open Weights License 0.X",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "LTX-Video update adding long-shot generation of up to 60 seconds of video, with new 13B and 2B distilled models from a shared base and an IC-LoRA detailer model.",
      "highlights": [
        "Up to 60 seconds of video",
        "13B and 2B distilled variants usable in one multiscale pipeline",
        "Compatible with the official IC-LoRA control models"
      ],
      "source_title": "Lightricks/LTX-Video",
      "source_url": "https://github.com/Lightricks/LTX-Video",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#ltx-video-0-9-8"
    },
    {
      "slug": "openreasoning-nemotron",
      "date": "2025-07-16",
      "model": "OpenReasoning-Nemotron",
      "lab": "NVIDIA",
      "lab_slug": "nvidia",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "1.5B, 7B, 14B and 32B",
      "model_license": "CC-BY-4.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Reasoning models derived from Qwen2.5 base models and post-trained for math, code and science solutions, released in four sizes under CC BY 4.0.",
      "highlights": [
        "32B model developed from Qwen2.5-32B",
        "Covers math, code and science reasoning",
        "Released under CC BY 4.0"
      ],
      "source_title": "nvidia/OpenReasoning-Nemotron-32B · Hugging Face",
      "source_url": "https://huggingface.co/nvidia/OpenReasoning-Nemotron-32B",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#openreasoning-nemotron"
    },
    {
      "slug": "cogvideox-3",
      "date": "2025-07-15",
      "model": "CogVideoX-3",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "CogVideoX-3 is Z.ai's upgraded video generation model on its API, supporting text-to-video and image-to-video. It adds first-and-last-frame generation and improves clarity and the smoothness of large subject motion.",
      "highlights": [
        "Adds start and end frame control",
        "Improved realistic and 3D-style scenes",
        "Available through the BigModel API platform"
      ],
      "source_title": "Model and product release records (BigModel docs)",
      "source_url": "https://docs.bigmodel.cn/cn/update/new-releases",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#cogvideox-3"
    },
    {
      "slug": "voxtral",
      "date": "2025-07-15",
      "model": "Voxtral Small and Voxtral Mini",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32000,
      "parameters": "24B and 3B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Mistral's first audio models, open-weight speech understanding models that combine transcription with question answering and summarization, in a 24B production size and a 3B edge size.",
      "highlights": [
        "Handles audio up to 30 minutes for transcription or 40 minutes for understanding",
        "The API also offers Voxtral Mini Transcribe for transcription",
        "Released under Apache 2.0"
      ],
      "source_title": "Voxtral | Mistral AI",
      "source_url": "https://mistral.ai/news/voxtral",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#voxtral"
    },
    {
      "slug": "kimi-k2",
      "date": "2025-07-11",
      "model": "Kimi K2",
      "lab": "Moonshot AI",
      "lab_slug": "moonshot-ai",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "MoE: 1T total, 32B active",
      "model_license": "Modified MIT License",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "Kimi K2 is Moonshot AI's open-weights mixture-of-experts model optimized for agentic tasks, released as Kimi-K2-Base for fine-tuning and Kimi-K2-Instruct for chat and agent use. Moonshot trained it with the MuonClip optimizer.",
      "highlights": [
        "1T total parameters with 32B activated",
        "128K context length at launch",
        "Offered on the Kimi Open Platform as kimi-k2-0711-preview",
        "Weights and code under the Modified MIT License"
      ],
      "source_title": "Kimi K2: Open Agentic Intelligence",
      "source_url": "https://www.kimi.com/en/blog/kimi-k2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#kimi-k2"
    },
    {
      "slug": "devstral-medium",
      "date": "2025-07-10",
      "model": "Devstral Medium",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "coding",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Enterprise"
      ],
      "summary": "Larger agentic coding model released through the API alongside Devstral Small 1.1, built with All Hands AI for software engineering agents.",
      "highlights": [
        "API model ID devstral-medium-2507",
        "Available for on-premises deployment and fine-tuning for enterprises",
        "Deprecated February 27, 2026"
      ],
      "source_title": "Upgrading agentic coding capabilities with the new Devstral models | Mistral AI",
      "source_url": "https://mistral.ai/news/devstral-2507",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#devstral-medium"
    },
    {
      "slug": "devstral-small-1-1",
      "date": "2025-07-10",
      "model": "Devstral Small 1.1",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "24B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Upgrade to Devstral Small that keeps the 24B architecture while improving generalization across prompts and agent scaffolds, built with All Hands AI.",
      "highlights": [
        "Supports both Mistral function calling and XML formats",
        "Released under Apache 2.0",
        "API model ID devstral-small-2507"
      ],
      "source_title": "Upgrading agentic coding capabilities with the new Devstral models | Mistral AI",
      "source_url": "https://mistral.ai/news/devstral-2507",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#devstral-small-1-1"
    },
    {
      "slug": "lfm2",
      "date": "2025-07-10",
      "model": "LFM2 (350M, 700M, 1.2B)",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32768,
      "parameters": "350M: 354M; 700M: 742M; 1.2B: 1.17B",
      "model_license": "LFM Open License v1.0",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "Second-generation Liquid Foundation Models using a new hybrid architecture of gated short convolutions and grouped query attention, released as open-weight 350M, 700M, and 1.2B dense checkpoints designed for fast on-device inference.",
      "highlights": [
        "Architecture: 10 double-gated short convolution blocks and 6 grouped query attention blocks",
        "Trained on 10T tokens with LFM1-7B as a knowledge-distillation teacher",
        "Weights on Hugging Face; also on Liquid Playground and OpenRouter",
        "LFM Open License v1.0 allows free commercial use below a company revenue threshold"
      ],
      "source_title": "Introducing LFM2: The Fastest On-Device Foundation Models on the Market",
      "source_url": "https://www.liquid.ai/blog/liquid-foundation-models-v2-our-second-series-of-generative-ai-models",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#lfm2"
    },
    {
      "slug": "reka-flash-3-1",
      "date": "2025-07-10",
      "model": "Reka Flash 3.1",
      "lab": "Reka",
      "lab_slug": "reka",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "21B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "An update to Reka Flash 3 from an upgraded reinforcement learning stack, aimed at coding and as a base for agentic fine-tuning, released with a 3.5-bit Reka Quant version and the Reka Quant quantization library.",
      "highlights": [
        "API model name reka-flash-3.1; also available in the Reka Playground",
        "3.5-bit Q3_K_S GGUF build quantized with Reka Quant for llama.cpp",
        "A customized version orchestrates Reka Research; a multimodal version underpins Reka Vision",
        "Weights on Hugging Face under Apache-2.0"
      ],
      "source_title": "Reka Flash 3.1 and Reka Quant",
      "source_url": "https://reka.ai/news/reka-flash-3-1-and-reka-quant",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#reka-flash-3-1"
    },
    {
      "slug": "flexolmo",
      "date": "2025-07-09",
      "model": "FlexOlmo",
      "lab": "Ai2",
      "lab_slug": "ai2",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Language model training paradigm in which each data owner branches from a shared public model and trains an expert locally on its own data, letting owners opt in or out later.",
      "highlights": [
        "Enables collaboration without pooling sensitive data",
        "Released as FlexOlmo-7x7B-1T on Hugging Face",
        "Released under Apache 2.0"
      ],
      "source_title": "Introducing FlexOlmo: a new paradigm for language model training and data collaboration | Ai2",
      "source_url": "https://allenai.org/blog/flexolmo",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#flexolmo"
    },
    {
      "slug": "grok-4",
      "date": "2025-07-09",
      "model": "Grok 4 and Grok 4 Heavy",
      "lab": "xAI",
      "lab_slug": "xai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Grok 4 is xAI's flagship trained with reinforcement learning at pretraining scale on the 200,000-GPU Colossus cluster, with native tool use and real-time search. Grok 4 Heavy adds parallel test-time compute.",
      "highlights": [
        "API offers a 256,000-token context window with text and vision understanding",
        "Available to SuperGrok and Premium+ subscribers and through the xAI API",
        "Grok 4 Heavy offered through a new SuperGrok Heavy tier",
        "API model grok-4-0709 retired May 15, 2026"
      ],
      "source_title": "Grok 4 | SpaceXSI",
      "source_url": "https://x.ai/news/grok-4",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#grok-4"
    },
    {
      "slug": "medgemma-27b-multimodal",
      "date": "2025-07-09",
      "model": "MedGemma 27B Multimodal",
      "lab": "Google",
      "lab_slug": "google",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "27B",
      "model_license": "Health AI Developer Foundations Terms of Use",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "MedGemma 27B Multimodal extends Google's open medical model collection with a 27B model that accepts images and text, adding longitudinal electronic health record interpretation. It was released together with the MedSigLIP medical image encoder.",
      "highlights": [
        "Complements the earlier MedGemma 4B Multimodal and 27B text models",
        "MedSigLIP is a roughly 400M-parameter SigLIP-based encoder for classification and retrieval",
        "All MedGemma models run on a single GPU"
      ],
      "source_title": "MedGemma: Our most capable open models for health AI development",
      "source_url": "https://research.google/blog/medgemma-our-most-capable-open-models-for-health-ai-development/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#medgemma-27b-multimodal"
    },
    {
      "slug": "phi-4-mini-flash-reasoning",
      "date": "2025-07-09",
      "model": "Phi-4-mini-flash-reasoning",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 64000,
      "parameters": "3.8B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights",
        "API",
        "Cloud platforms"
      ],
      "summary": "Phi-4-mini-flash-reasoning is a 3.8B open Microsoft model for math reasoning on edge and mobile devices, built on the new SambaY decoder-hybrid-decoder architecture that combines Mamba, sliding window attention and gated memory units.",
      "highlights": [
        "Uses the SambaY architecture with a Gated Memory Unit",
        "Supports a 64K token context length",
        "Available on Azure AI Foundry, NVIDIA API Catalog and Hugging Face"
      ],
      "source_title": "Reasoning reimagined: Introducing Phi-4-mini-flash-reasoning",
      "source_url": "https://azure.microsoft.com/en-us/blog/reasoning-reimagined-introducing-phi-4-mini-flash-reasoning/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#phi-4-mini-flash-reasoning"
    },
    {
      "slug": "t5gemma",
      "date": "2025-07-09",
      "model": "T5Gemma",
      "lab": "Google",
      "lab_slug": "google",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "2B-2B, 9B-9B, 9B-2B, plus T5-sized Small to XL",
      "model_license": "Gemma Terms of Use",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "T5Gemma is a collection of encoder-decoder language models created by adapting pretrained Gemma 2 decoder-only models and further training them with UL2 or PrefixLM objectives. It includes balanced and unbalanced encoder-decoder pairings.",
      "highlights": [
        "Gemma 2-based configurations include 2B-2B, 9B-9B and an unbalanced 9B-2B",
        "Pretrained and instruction-tuned checkpoints",
        "Weights on Hugging Face and Kaggle; inference on Vertex AI"
      ],
      "source_title": "T5Gemma: A new collection of encoder-decoder Gemma models",
      "source_url": "https://developers.googleblog.com/en/t5gemma/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#t5gemma"
    },
    {
      "slug": "smollm3",
      "date": "2025-07-08",
      "model": "SmolLM3",
      "lab": "Hugging Face",
      "lab_slug": "hugging-face",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "3B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Fully open 3B model with dual-mode reasoning, six languages and long context, released with an engineering blueprint covering its architecture and data mixtures.",
      "highlights": [
        "Trained on 11 trillion tokens",
        "Trained at 64K context and extends to 128K with YaRN",
        "Supports English, French, Spanish, German, Italian and Portuguese"
      ],
      "source_title": "SmolLM3: smol, multilingual, long-context reasoner",
      "source_url": "https://huggingface.co/blog/smollm3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#smollm3"
    },
    {
      "slug": "jamba-1-7",
      "date": "2025-07-03",
      "model": "Jamba 1.7",
      "lab": "AI21 Labs",
      "lab_slug": "ai21",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Jamba Open Model License",
      "available_at_launch": [
        "Open weights",
        "API",
        "Enterprise"
      ],
      "summary": "Update to Jamba Large and Jamba Mini with better grounding for more complete and accurate answers, plus optimized configurations and quantization.",
      "highlights": [
        "Available for self-hosted deployment",
        "Weights on Hugging Face as Jamba Large 1.7 and Jamba Mini 1.7",
        "Suited to question answering and instruction following"
      ],
      "source_title": "Changelog - AI21 Labs",
      "source_url": "https://docs.ai21.com/changelog",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#jamba-1-7"
    },
    {
      "slug": "glm-4-1v-thinking",
      "date": "2025-07-01",
      "model": "GLM-4.1V-9B-Thinking",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 64000,
      "parameters": "9B",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "GLM-4.1V-9B-Thinking is Z.ai's open-weights vision-language reasoning model, built on GLM-4-9B-0414. It targets video understanding, image question answering, chart reading, and GUI operation with step-by-step reasoning.",
      "highlights": [
        "Released with the GLM-4.1V-9B-Base model",
        "64K context length support",
        "Weights under the MIT license; API launched July 2, 2025"
      ],
      "source_title": "GLM-V (GitHub)",
      "source_url": "https://github.com/zai-org/GLM-V",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/07#glm-4-1v-thinking"
    },
    {
      "slug": "ernie-4-5-open-weights",
      "date": "2025-06-30",
      "model": "ERNIE 4.5 (open-weight family)",
      "lab": "Baidu",
      "lab_slug": "baidu",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 131072,
      "parameters": "MoE: up to 424B total, 47B active; plus a 0.3B dense model",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Baidu open-sourced the ERNIE 4.5 family of 10 variants: mixture-of-experts language and vision-language models with 47B or 3B active parameters plus a 0.3B dense model, trained with PaddlePaddle and offered in PyTorch-compatible formats.",
      "highlights": [
        "Largest models are ERNIE-4.5-300B-A47B (text) and ERNIE-4.5-VL-424B-A47B (vision-language)",
        "131,072-token context length per the Hugging Face model cards",
        "Vision-language models support thinking and non-thinking modes",
        "Released with the ERNIEKit fine-tuning and FastDeploy deployment toolkits"
      ],
      "source_title": "Announcing the Open Source Release of the ERNIE 4.5 Model Family",
      "source_url": "https://ernie.baidu.com/research/posts/ernie4.5/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/06#ernie-4-5-open-weights"
    },
    {
      "slug": "seed-1-6-embedding",
      "date": "2025-06-28",
      "model": "Seed1.6-Embedding",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "embedding",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Seed1.6-Embedding is a multimodal embedding model built on Seed1.6-Flash that supports hybrid retrieval across text, image, and video, with selectable 2048 or 1024 embedding dimensions.",
      "highlights": [
        "API model ID doubao-embedding-vision-250615 on Volcano Engine",
        "Dual-tower design using the final-layer EOS token representation",
        "Supports 2048 and 1024 embedding dimensions"
      ],
      "source_title": "Built on Seed1.6-Flash, Seed-1.6-Embedding Launched",
      "source_url": "https://seed.bytedance.com/en/blog/built-on-seed1-6-flash-seed-1-6-embedding-launched",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/06#seed-1-6-embedding"
    },
    {
      "slug": "hunyuan-a13b",
      "date": "2025-06-27",
      "model": "Hunyuan-A13B",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "MoE: 80B total, 13B active",
      "model_license": "Tencent Hunyuan Community License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Hunyuan-A13B is an open fine-grained mixture-of-experts language model with 80B total and 13B active parameters, supporting hybrid fast and slow thinking for reasoning and agent tasks in resource-constrained environments.",
      "highlights": [
        "Native 256K context window",
        "Pretrain, Instruct, FP8, and GPTQ-Int4 checkpoints released with a technical report",
        "Defaults to slow-thinking mode, which can be turned off"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/Hunyuan-A13B: Tencent Hunyuan A13B (short as Hunyuan-A13B), an innovative and open-source LLM built on a fine-grained MoE architecture.",
      "source_url": "https://github.com/Tencent-Hunyuan/Hunyuan-A13B",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/06#hunyuan-a13b"
    },
    {
      "slug": "qwen-tts",
      "date": "2025-06-27",
      "model": "Qwen-TTS",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Qwen-TTS is Qwen's text-to-speech model on the Qwen API, trained on millions of hours of speech. This update added support for several Chinese dialects.",
      "highlights": [
        "Model IDs qwen-tts-latest and qwen-tts-2025-05-22",
        "Generates 3 Chinese dialects: Pekingese, Shanghainese, and Sichuanese",
        "Available through the Qwen API"
      ],
      "source_title": "Time to Speak Some Dialects, Qwen-TTS!",
      "source_url": "https://qwenlm.github.io/blog/qwen-tts/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/06#qwen-tts"
    },
    {
      "slug": "flux-1-kontext-dev",
      "date": "2025-06-26",
      "model": "FLUX.1 Kontext [dev]",
      "lab": "Black Forest Labs",
      "lab_slug": "black-forest-labs",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "12B",
      "model_license": "FLUX.1 [dev] Non-Commercial License",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "Open-weight developer version of FLUX.1 Kontext focused exclusively on image editing, a 12B-parameter model that runs on consumer hardware and supports iterative, local and global edits.",
      "highlights": [
        "12B parameters, compatible with FLUX.1 [dev] inference code",
        "BF16, FP8 and FP4 TensorRT variants for NVIDIA Blackwell",
        "Released alongside the KontextBench evaluation set and a technical report"
      ],
      "source_title": "FLUX.1 Kontext [dev] - Open Weights for Image Editing",
      "source_url": "https://bfl.ai/blog/flux-1-kontext-dev",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/06#flux-1-kontext-dev"
    },
    {
      "slug": "keye-vl-8b-preview",
      "date": "2025-06-26",
      "model": "Keye-VL-8B-Preview",
      "lab": "Kuaishou",
      "lab_slug": "kuaishou",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "8B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Keye-VL is Kuaishou's first open multimodal large language model, built to understand short-form video as well as images, with a switchable thinking mode for reasoning.",
      "highlights": [
        "Focused on dynamic, information-dense short-form video",
        "Soft switch between thinking and non-thinking modes",
        "Technical report followed on July 1, 2025"
      ],
      "source_title": "GitHub - Kwai-Keye/Keye",
      "source_url": "https://github.com/Kwai-Keye/Keye",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/06#keye-vl-8b-preview"
    },
    {
      "slug": "o3-deep-research",
      "date": "2025-06-26",
      "model": "o3-deep-research and o4-mini-deep-research",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 200000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "o3-deep-research and o4-mini-deep-research bring the deep research models used in ChatGPT to the API, letting developers trigger multi-step research that can pull external context through web search and MCP servers.",
      "highlights": [
        "Snapshots o3-deep-research-2025-06-26 and o4-mini-deep-research-2025-06-26",
        "Launched with API webhooks; OpenAI recommends background mode for long tasks",
        "200,000-token context window per the API model pages"
      ],
      "source_title": "Deep research in the API, webhooks, and web search with o3 - Announcements - OpenAI Developer Community",
      "source_url": "https://community.openai.com/t/deep-research-in-the-api-webhooks-and-web-search-with-o3/1299919",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/06#o3-deep-research"
    },
    {
      "slug": "qwen-vlo",
      "date": "2025-06-26",
      "model": "Qwen VLo",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Qwen VLo is a preview unified multimodal understanding and generation model from Qwen, which generates high-quality image recreations based on its understanding of the input.",
      "highlights": [
        "Preview version available through Qwen Chat",
        "Supports instructions in multiple languages, including Chinese and English",
        "Extends Qwen's vision models from understanding the world to depicting it"
      ],
      "source_title": "Qwen VLo: From Understanding the World to Depicting It",
      "source_url": "https://qwenlm.github.io/blog/qwen-vlo/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/06#qwen-vlo"
    },
    {
      "slug": "seed-1-6",
      "date": "2025-06-25",
      "model": "Seed1.6",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "MoE: 230B total, 23B active",
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Seed1.6 is ByteDance Seed's general-purpose model series combining multimodal understanding, adaptive deep thinking, GUI interaction, and a 256K context window, offered on Volcano Engine with Seed1.6-Thinking and Seed1.6-Flash variants.",
      "highlights": [
        "Sparse MoE pretrained with 23B active and 230B total parameters",
        "Adaptive chain-of-thought (AdaCoT) decides when to think based on difficulty",
        "Long-context training raised sequence length from 32K to 256K",
        "Listed in Volcano Engine's June 2025 model release notes as doubao-seed-1-6-250615 with thinking and flash variants"
      ],
      "source_title": "Introduction to Techniques Used in Seed1.6",
      "source_url": "https://seed.bytedance.com/en/blog/introduction-to-techniques-used-in-seed1-6",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/06#seed-1-6"
    },
    {
      "slug": "eleven-voice-design-v3",
      "date": "2025-06-25",
      "model": "Voice Design v3",
      "lab": "ElevenLabs",
      "lab_slug": "elevenlabs",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Text-to-voice model that creates custom voices from a written description, returning three options per prompt for immediate use in projects.",
      "highlights": [
        "Model ID eleven_ttv_v3",
        "Two design modes",
        "Generates three voice options per description"
      ],
      "source_title": "Voice Design v3",
      "source_url": "https://elevenlabs.io/blog/voice-design-v3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/06#eleven-voice-design-v3"
    },
    {
      "slug": "kimi-vl-a3b-thinking-2506",
      "date": "2025-06-21",
      "model": "Kimi-VL-A3B-Thinking-2506",
      "lab": "Moonshot AI",
      "lab_slug": "moonshot-ai",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "MoE: 16B total, 3B active",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Kimi-VL-A3B-Thinking-2506 is an improved version of Moonshot AI's Kimi-VL-A3B-Thinking vision-language reasoning model. Moonshot recommends using it instead of the earlier Thinking checkpoint released in April 2025.",
      "highlights": [
        "Released with a tech blog and cookbook on Hugging Face",
        "Same 16B-total, 3B-active mixture-of-experts design as Kimi-VL",
        "Weights under the MIT License"
      ],
      "source_title": "Kimi-VL (GitHub)",
      "source_url": "https://github.com/MoonshotAI/Kimi-VL",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/06#kimi-vl-a3b-thinking-2506"
    },
    {
      "slug": "magenta-realtime",
      "date": "2025-06-20",
      "model": "Magenta RealTime",
      "lab": "Google",
      "lab_slug": "google",
      "type": "music",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": "800M",
      "model_license": "CC-BY-4.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Magenta RealTime is an open-weights live music model from Google's Magenta team that generates music in two-second chunks steered by text or audio prompts. It is the open counterpart to the Lyria RealTime API.",
      "highlights": [
        "800M-parameter autoregressive transformer",
        "Released as a research preview",
        "Weights on Hugging Face and Google Cloud Storage; weights under CC BY 4.0, code under Apache 2.0"
      ],
      "source_title": "Magenta RealTime: An Open-Weights Live Music Model",
      "source_url": "https://magenta.withgoogle.com/magenta-realtime",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/06#magenta-realtime"
    },
    {
      "slug": "minimax-music-1-5",
      "date": "2025-06-20",
      "model": "MiniMax Music 1.5",
      "lab": "MiniMax",
      "lab_slug": "minimax",
      "type": "music",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "MiniMax Music 1.5 is a music generation model that creates songs from style prompts and lyrics, extending generation to full four-minute tracks with more natural vocals, layered instrumentals, and coherent song structure.",
      "highlights": [
        "Beta listed in the API release notes on June 20, 2025",
        "Full release with 4-minute songs announced September 11, 2025",
        "Advanced mode assigns lyrics to sections such as Intro, Verse, and Chorus"
      ],
      "source_title": "Introducing MiniMax Music 1.5: Craft 4-Minute Masterpieces and Rediscover the Soul of Music",
      "source_url": "https://www.minimax.io/news/minimax-music-15",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/06#minimax-music-1-5"
    },
    {
      "slug": "mistral-small-3-2",
      "date": "2025-06-20",
      "model": "Mistral Small 3.2",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "24B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Minor update of Mistral Small 3.1 with a more robust function calling template, released as open weights and added to the API three days later.",
      "highlights": [
        "API model ID mistral-small-2506, available June 23, 2025",
        "Accepts text and image input",
        "Deprecated April 30, 2026 in favor of Mistral Small 4"
      ],
      "source_title": "Mistral Small 3.2 | Mistral Docs",
      "source_url": "https://docs.mistral.ai/models/mistral-small-3-2-25-06",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/06#mistral-small-3-2"
    },
    {
      "slug": "midjourney-video-v1",
      "date": "2025-06-18",
      "model": "Midjourney Video V1",
      "lab": "Midjourney",
      "lab_slug": "midjourney",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "image",
        "text"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Midjourney's first video model, an image-to-video workflow that animates Midjourney images with automatic or manual motion prompts and low or high motion settings.",
      "highlights": [
        "Each job produces four 5-second videos",
        "Videos can be extended about 4 seconds at a time, four times",
        "Launched web-only; HD 720p mode added August 2025"
      ],
      "source_title": "Introducing Our V1 Video Model",
      "source_url": "https://updates.midjourney.com/introducing-our-v1-video-model/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/06#midjourney-video-v1"
    },
    {
      "slug": "minimax-hailuo-02",
      "date": "2025-06-18",
      "model": "MiniMax Hailuo 02",
      "lab": "MiniMax",
      "lab_slug": "minimax",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "MiniMax Hailuo 02 is a video generation model with native 1080p output, stronger instruction following, and better rendering of complex physics, built on a new architecture with three times the parameters of its predecessor.",
      "highlights": [
        "Offered in 768p-6s, 768p-10s, and 1080p-6s versions",
        "Trained on four times the data of the previous model",
        "Available on Hailuo Video web and mobile and the MiniMax API"
      ],
      "source_title": "MiniMax Hailuo 02, World-Class Quality, Record-Breaking Cost Efficiency",
      "source_url": "https://www.minimax.io/news/minimax-hailuo-02",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/06#minimax-hailuo-02"
    },
    {
      "slug": "gemini-2-5-flash-lite",
      "date": "2025-06-17",
      "model": "Gemini 2.5 Flash-Lite",
      "lab": "Google",
      "lab_slug": "google",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1048576,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms"
      ],
      "summary": "Gemini 2.5 Flash-Lite is the most cost-efficient and fastest model in the Gemini 2.5 family, aimed at high-volume, latency-sensitive tasks such as translation and classification.",
      "highlights": [
        "Preview released June 17, 2025, the same day 2.5 Pro and 2.5 Flash became generally available",
        "Stable gemini-2.5-flash-lite became generally available on July 22, 2025",
        "1 million token context window, controllable thinking budgets and native tools such as Grounding with Google Search"
      ],
      "source_title": "Gemini 2.5 model family expands",
      "source_url": "https://blog.google/products-and-platforms/products/gemini/gemini-2-5-model-family-expands/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/06#gemini-2-5-flash-lite"
    },
    {
      "slug": "kimi-dev-72b",
      "date": "2025-06-17",
      "model": "Kimi-Dev-72B",
      "lab": "Moonshot AI",
      "lab_slug": "moonshot-ai",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "72B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Kimi-Dev-72B is Moonshot AI's open-weights coding model for software engineering issue resolution. It was optimized with large-scale reinforcement learning in which it patches real repositories in Docker and is rewarded only when the full test suite passes.",
      "highlights": [
        "Targets SWE-bench-style repository issue resolution",
        "Available on Hugging Face and GitHub",
        "Weights under the MIT License"
      ],
      "source_title": "moonshotai/Kimi-Dev-72B (Hugging Face)",
      "source_url": "https://huggingface.co/moonshotai/Kimi-Dev-72B",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/06#kimi-dev-72b"
    },
    {
      "slug": "minimax-m1",
      "date": "2025-06-16",
      "model": "MiniMax-M1",
      "lab": "MiniMax",
      "lab_slug": "minimax",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": "MoE: 456B total, 45.9B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "MiniMax-M1 is an open-weight hybrid-attention reasoning model built on MiniMax-Text-01, combining Lightning Attention with mixture-of-experts and trained with large-scale reinforcement learning using MiniMax's CISPO algorithm.",
      "highlights": [
        "Native 1M-token context with up to 80K tokens of reasoning output",
        "Released in 40K and 80K thinking-budget versions",
        "Weights under the Apache-2.0 license",
        "Weights and technical report published on Hugging Face and GitHub"
      ],
      "source_title": "MiniMax-M1, the World's First Open-Source, Large-Scale, Hybrid-Attention Reasoning Model",
      "source_url": "https://www.minimax.io/news/minimaxm1",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/06#minimax-m1"
    },
    {
      "slug": "hunyuan3d-2-1",
      "date": "2025-06-13",
      "model": "Hunyuan3D 2.1",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "image"
      ],
      "output_modalities": [
        "3d"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Tencent Hunyuan 3D 2.1 Community License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Hunyuan3D 2.1 is a fully open 3D asset generation system that releases full model weights and training code, replacing the earlier RGB texture model with a physically based rendering (PBR) material pipeline.",
      "highlights": [
        "Includes image-to-shape and PBR texture generation models",
        "Technical report published June 19, 2025"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/Hunyuan3D-2.1: From Images to High-Fidelity 3D Assets with Production-Ready PBR Material",
      "source_url": "https://github.com/Tencent-Hunyuan/Hunyuan3D-2.1",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/06#hunyuan3d-2-1"
    },
    {
      "slug": "granite-vision-3-3",
      "date": "2025-06-11",
      "model": "Granite Vision 3.3",
      "lab": "IBM",
      "lab_slug": "ibm",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "2B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Compact IBM vision-language model for visual document understanding that automates content extraction from tables, charts, infographics, plots and diagrams, succeeding Granite Vision 3.2.",
      "highlights": [
        "Released with a Granite Vision 3.3 2B embedding variant",
        "Successor to Granite Vision 3.2",
        "Released under Apache 2.0"
      ],
      "source_title": "ibm-granite/granite-vision-3.3-2b · Hugging Face",
      "source_url": "https://huggingface.co/ibm-granite/granite-vision-3.3-2b",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/06#granite-vision-3-3"
    },
    {
      "slug": "seedance-1-0",
      "date": "2025-06-11",
      "model": "Seedance 1.0",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Seedance 1.0 is ByteDance Seed's foundation video generation model, producing 1080p video from text and image prompts with native multi-shot storytelling, stable motion, and fast inference.",
      "highlights": [
        "Generates 10-second clips with 2 to 3 shot transitions",
        "Available through Jimeng, Doubao, and the Volcano Engine API",
        "Volcano Engine offers Seedance 1.0 Pro and Lite versions",
        "Technical report published at launch"
      ],
      "source_title": "Tech Report of Seedance 1.0 Is Now Publicly Available",
      "source_url": "https://seed.bytedance.com/en/blog/tech-report-of-seedance-1-0-is-now-publicly-available",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/06#seedance-1-0"
    },
    {
      "slug": "magistral-medium",
      "date": "2025-06-10",
      "model": "Magistral Medium",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 40000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Enterprise version of Magistral, Mistral AI's first reasoning model, focused on domain-specific, transparent and multilingual reasoning for use across professional domains.",
      "highlights": [
        "Released alongside the open Magistral Small",
        "Previewed in Le Chat",
        "API model ID magistral-medium-2506"
      ],
      "source_title": "Magistral | Mistral AI",
      "source_url": "https://mistral.ai/news/magistral",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/06#magistral-medium"
    },
    {
      "slug": "magistral-small",
      "date": "2025-06-10",
      "model": "Magistral Small",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 40000,
      "parameters": "24B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Open-weight version of Magistral, Mistral AI's first reasoning model, built for transparent, multilingual step-by-step reasoning that users can follow and verify.",
      "highlights": [
        "Released alongside the enterprise Magistral Medium",
        "Weights published as Magistral-Small-2506",
        "API model ID magistral-small-2506"
      ],
      "source_title": "Magistral | Mistral AI",
      "source_url": "https://mistral.ai/news/magistral",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/06#magistral-small"
    },
    {
      "slug": "o3-pro",
      "date": "2025-06-10",
      "model": "OpenAI o3-pro",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 200000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "o3-pro is a version of OpenAI o3 designed to think longer and give more reliable answers. It replaced o1-pro for ChatGPT Pro and Team users and launched in the API the same day.",
      "highlights": [
        "API model ID o3-pro (snapshot o3-pro-2025-06-10)",
        "Has ChatGPT tools including web search, file analysis, Python, and memory",
        "Enterprise and Edu ChatGPT access followed the week after launch",
        "200,000-token context window per the API model page"
      ],
      "source_title": "Model Release Notes | OpenAI Help Center",
      "source_url": "https://help.openai.com/en/articles/9624314-model-release-notes",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/06#o3-pro"
    },
    {
      "slug": "apple-foundation-models-2025",
      "date": "2025-06-09",
      "model": "Apple Foundation Models (2025)",
      "lab": "Apple",
      "lab_slug": "apple",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "About 3B (on-device); server size not disclosed",
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Apple's 2025 foundation models update an on-device model of about 3B parameters and a server model built on a new parallel track mixture-of-experts architecture, adding image understanding and 15 languages.",
      "highlights": [
        "Foundation Models framework gives developers direct access to the on-device model",
        "Framework supports guided generation via Swift @Generable and tool calling",
        "Server model runs on Private Cloud Compute",
        "Vision encoders: 300M ViTDet-L on device and 1B ViT-g on the server"
      ],
      "source_title": "Updates to Apple's On-Device and Server Foundation Language Models",
      "source_url": "https://machinelearning.apple.com/research/apple-foundation-models-2025-updates",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/06#apple-foundation-models-2025"
    },
    {
      "slug": "seededit-3-0",
      "date": "2025-06-06",
      "model": "SeedEdit 3.0",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "SeedEdit 3.0 is an instruction-based image editing model built on Seedream 3.0 that better preserves subjects, backgrounds, and details, suited to portrait retouching, background changes, perspective shifts, and lighting adjustments.",
      "highlights": [
        "Processes and outputs images above 1K resolution",
        "Opened for testing on the Jimeng web app at launch",
        "Volcano Engine later added it as doubao-seededit-3.0-i2i"
      ],
      "source_title": "ByteDance Released Image Editing Model SeedEdit 3.0: Enhanced Image Consistency and Usability Rate",
      "source_url": "https://seed.bytedance.com/en/blog/bytedance-released-image-editing-model-seededit-3-0-enhanced-image-consistency-and-usability-rate",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/06#seededit-3-0"
    },
    {
      "slug": "qwen3-embedding",
      "date": "2025-06-05",
      "model": "Qwen3-Embedding",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "embedding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32000,
      "parameters": "0.6B, 4B, 8B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Qwen3-Embedding is a series of text embedding models built on Qwen3 foundation models, released in 0.6B, 4B, and 8B sizes for retrieval, clustering, and classification.",
      "highlights": [
        "Released together with the Qwen3-Reranker series",
        "32K context length",
        "Weights on Hugging Face and ModelScope under the Apache 2.0 license"
      ],
      "source_title": "Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Models",
      "source_url": "https://qwenlm.github.io/blog/qwen3-embedding/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/06#qwen3-embedding"
    },
    {
      "slug": "qwen3-reranker",
      "date": "2025-06-05",
      "model": "Qwen3-Reranker",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "reranker",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32000,
      "parameters": "0.6B, 4B, 8B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Qwen3-Reranker is a series of text reranking models built on Qwen3 foundation models, released alongside Qwen3-Embedding for retrieval pipelines that rerank candidates found by dense embedding search.",
      "highlights": [
        "Available in 0.6B, 4B, and 8B sizes",
        "32K context length",
        "Weights under the Apache 2.0 license"
      ],
      "source_title": "Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Models",
      "source_url": "https://qwenlm.github.io/blog/qwen3-embedding/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/06#qwen3-reranker"
    },
    {
      "slug": "eleven-v3",
      "date": "2025-06-03",
      "model": "Eleven v3",
      "lab": "ElevenLabs",
      "lab_slug": "elevenlabs",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Expressive text-to-speech model released as a research preview alpha, supporting 70+ languages, multi-speaker dialogue and inline audio tags such as [excited], [whispers] and [sighs].",
      "highlights": [
        "Added to the API on August 20, 2025",
        "Generally available from February 2, 2026",
        "Dialogue mode with multiple speakers"
      ],
      "source_title": "Eleven v3 (alpha)",
      "source_url": "https://elevenlabs.io/blog/eleven-v3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/06#eleven-v3"
    },
    {
      "slug": "llama-nemotron-nano-vl",
      "date": "2025-06-03",
      "model": "Llama Nemotron Nano VL 8B",
      "lab": "NVIDIA",
      "lab_slug": "nvidia",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "8B",
      "model_license": "NVIDIA Open Model License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Document intelligence vision-language model that queries and summarizes images, deployable in the data center, in the cloud and at the edge.",
      "highlights": [
        "Focused on document understanding",
        "Part of the Llama Nemotron family",
        "Released under the NVIDIA Open Model License"
      ],
      "source_title": "nvidia/Llama-3.1-Nemotron-Nano-VL-8B-V1 · Hugging Face",
      "source_url": "https://huggingface.co/nvidia/Llama-3.1-Nemotron-Nano-VL-8B-V1",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/06#llama-nemotron-nano-vl"
    },
    {
      "slug": "leonardo-lucid-realism",
      "date": "2025-06-03",
      "model": "Lucid Realism",
      "lab": "Leonardo.Ai",
      "lab_slug": "leonardo-ai",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Leonardo.Ai image model for cinematic, story-driven visuals, trained with image-to-video generation in mind and suited to start frames for Motion 2.0 video.",
      "highlights": [
        "Aimed at photo-like fashion shoots, interiors, lifestyle and ads",
        "Built to pair with Motion 2.0 video workflows"
      ],
      "source_title": "Leonardo.Ai on X: For Creators Who Think in Frames. Lucid Realism is our latest model built for cinematic visuals",
      "source_url": "https://x.com/LeonardoAi/status/1930046253636657397",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/06#leonardo-lucid-realism"
    },
    {
      "slug": "mimo-vl-7b",
      "date": "2025-05-30",
      "model": "MiMo-VL-7B",
      "lab": "Xiaomi",
      "lab_slug": "xiaomi",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "7B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "MiMo-VL-7B is Xiaomi's open-source vision-language model series, released as SFT and RL checkpoints, with the RL version trained through Mixed On-policy Reinforcement Learning across perception, grounding, reasoning, and preference rewards.",
      "highlights": [
        "Four-stage pre-training ending in long-context supervised fine-tuning",
        "MiMo-VL-7B-2508 update in August 2025 added a switch to turn reasoning off",
        "Weights under the MIT license"
      ],
      "source_title": "XiaomiMiMo/MiMo-VL",
      "source_url": "https://github.com/XiaomiMiMo/MiMo-VL",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/05#mimo-vl-7b"
    },
    {
      "slug": "flux-1-kontext-pro-max",
      "date": "2025-05-29",
      "model": "FLUX.1 Kontext [pro] and [max]",
      "lab": "Black Forest Labs",
      "lab_slug": "black-forest-labs",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Cloud platforms"
      ],
      "summary": "In-context image generation and editing models that accept text and reference images, enabling character consistency, local edits and style reference, plus multi-turn iterative editing; [max] adds stronger prompt adherence and typography.",
      "highlights": [
        "Unifies text-to-image generation and instruction-based editing",
        "Launched with the BFL Playground",
        "Available via the BFL API and partners such as fal, Replicate and Together AI"
      ],
      "source_title": "Introducing FLUX.1 Kontext and the BFL Playground",
      "source_url": "https://bfl.ai/blog/flux-1-kontext",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/05#flux-1-kontext-pro-max"
    },
    {
      "slug": "codestral-embed",
      "date": "2025-05-28",
      "model": "Codestral Embed",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "embedding",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "embedding"
      ],
      "context_window": 8000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Mistral's first embedding model specialized for code, built for high-performance retrieval and semantic understanding over real-world code data in coding assistants and agents.",
      "highlights": [
        "Outputs embeddings at multiple dimensions and precisions",
        "Dimensions are ordered by relevance, so vectors can be truncated",
        "API model ID codestral-embed-2505"
      ],
      "source_title": "Codestral Embed | Mistral AI",
      "source_url": "https://mistral.ai/news/codestral-embed",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/05#codestral-embed"
    },
    {
      "slug": "deepseek-r1-0528",
      "date": "2025-05-28",
      "model": "DeepSeek-R1-0528",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "DeepSeek-R1-0528 is an update to DeepSeek-R1 with improved reasoning, enhanced front-end generation, and reduced hallucinations. It added JSON output and function calling support to the reasoning model.",
      "highlights": [
        "Replaced R1 behind the deepseek-reasoner API model with no change to API usage",
        "Released with DeepSeek-R1-0528-Qwen3-8B, a distilled 8B model",
        "Weights on Hugging Face under the MIT License"
      ],
      "source_title": "DeepSeek-R1-0528 Release",
      "source_url": "https://api-docs.deepseek.com/news/news250528",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/05#deepseek-r1-0528"
    },
    {
      "slug": "hunyuanvideo-avatar",
      "date": "2025-05-28",
      "model": "HunyuanVideo-Avatar",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "image",
        "audio"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Tencent Hunyuan Community License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "HunyuanVideo-Avatar is an open audio-driven human animation model based on a multimodal diffusion transformer that turns avatar images into dynamic, emotion-controllable videos, including multi-character dialogue.",
      "highlights": [
        "Supports photorealistic, cartoon, 3D-rendered, and anthropomorphic characters",
        "Portrait, upper-body, and full-body generation"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/HunyuanVideo-Avatar",
      "source_url": "https://github.com/Tencent-Hunyuan/HunyuanVideo-Avatar",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/05#hunyuanvideo-avatar"
    },
    {
      "slug": "kling-2-1",
      "date": "2025-05-26",
      "model": "Kling 2.1",
      "lab": "Kuaishou",
      "lab_slug": "kuaishou",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Kling 2.1 is a model series that adds Standard and High Quality video generation modes alongside a premium Kling 2.1 Master edition with stronger motion and prompt responsiveness.",
      "highlights": [
        "Standard (720p) and High Quality (1080p) modes",
        "Kling 2.1 Master offers improved motion and semantic responsiveness",
        "Start and end frame control for Kling 2.1 arrived on August 21, 2025"
      ],
      "source_title": "Kling AI Release Notes & Version History | Kling AI",
      "source_url": "https://kling.ai/release-note/release-history",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/05#kling-2-1"
    },
    {
      "slug": "claude-4",
      "date": "2025-05-22",
      "model": "Claude Opus 4 and Claude Sonnet 4",
      "lab": "Anthropic",
      "lab_slug": "anthropic",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Cloud platforms",
        "Enterprise"
      ],
      "summary": "Claude Opus 4 and Claude Sonnet 4 are Anthropic's fourth-generation hybrid models, offering near-instant responses or extended thinking, with Opus 4 aimed at long-running coding and agent workflows and Sonnet 4 succeeding Claude 3.7 Sonnet.",
      "highlights": [
        "API model IDs claude-opus-4-20250514 and claude-sonnet-4-20250514",
        "Extended thinking with tool use in beta; both models can use tools in parallel",
        "Opus 4 deployed under the ASL-3 Standard and Sonnet 4 under ASL-2",
        "Sonnet 4 gained a 1M-token context window in beta on August 12, 2025"
      ],
      "source_title": "Introducing Claude 4",
      "source_url": "https://www.anthropic.com/news/claude-4",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/05#claude-4"
    },
    {
      "slug": "mistral-ocr-2",
      "date": "2025-05-22",
      "model": "Mistral OCR 2",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Second-generation Mistral OCR service released together with document annotations, powering Mistral's Document AI stack so users can extract interleaved text and images from documents.",
      "highlights": [
        "API model ID mistral-ocr-2505",
        "Released with the annotations feature",
        "Deprecated February 27, 2026"
      ],
      "source_title": "Changelog | Mistral Docs",
      "source_url": "https://docs.mistral.ai/resources/changelogs",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/05#mistral-ocr-2"
    },
    {
      "slug": "devstral",
      "date": "2025-05-21",
      "model": "Devstral",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "24B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Agentic coding model built with All Hands AI and trained to resolve real GitHub issues inside agent scaffolds such as OpenHands and SWE-Agent.",
      "highlights": [
        "Light enough for a single RTX 4090 or a Mac with 32GB of RAM",
        "Fine-tuned from Mistral Small 3.1",
        "API model ID devstral-small-2505"
      ],
      "source_title": "Devstral | Mistral AI",
      "source_url": "https://mistral.ai/news/devstral",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/05#devstral"
    },
    {
      "slug": "bytedance-bagel",
      "date": "2025-05-20",
      "model": "BAGEL",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text",
        "image"
      ],
      "context_window": null,
      "parameters": "MoT: 14B total, 7B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "BAGEL is an open-source unified multimodal model from ByteDance Seed that handles understanding, image generation, and image editing in one Mixture-of-Transformer-Experts architecture pretrained on interleaved text, image, video, and web data.",
      "highlights": [
        "7B active and 14B total parameters",
        "Fine-tuned from Qwen2.5-7B-Instruct and SigLIP",
        "Website, demo, weights, and report released on May 20, 2025"
      ],
      "source_title": "GitHub - ByteDance-Seed/Bagel: Open-source unified multimodal model",
      "source_url": "https://github.com/ByteDance-Seed/Bagel",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/05#bytedance-bagel"
    },
    {
      "slug": "gemini-2-5-flash-native-audio",
      "date": "2025-05-20",
      "model": "Gemini 2.5 Flash Native Audio",
      "lab": "Google",
      "lab_slug": "google",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio",
        "video",
        "text"
      ],
      "output_modalities": [
        "audio",
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Gemini 2.5 Flash native audio models brought native audio output to the Live API, enabling real-time spoken dialogue with natural voices, proactive audio that ignores background speech, and support for 24+ languages.",
      "highlights": [
        "Preview models gemini-2.5-flash-preview-native-audio-dialog and gemini-2.5-flash-exp-native-audio-thinking-dialog",
        "Announced at Google I/O 2025 with the Gemini 2.5 updates",
        "Updated native audio previews followed in September and December 2025"
      ],
      "source_title": "Google I/O 2025: Updates to Gemini 2.5 from Google DeepMind",
      "source_url": "https://blog.google/innovation-and-ai/models-and-research/google-deepmind/google-gemini-updates-io-2025/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/05#gemini-2-5-flash-native-audio"
    },
    {
      "slug": "gemini-2-5-tts",
      "date": "2025-05-20",
      "model": "Gemini 2.5 Flash TTS and Gemini 2.5 Pro TTS",
      "lab": "Google",
      "lab_slug": "google",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": 8192,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Gemini 2.5 Flash TTS and Gemini 2.5 Pro TTS are controllable text-to-speech preview models that generate speech with one or two speakers, letting developers direct style, tone and pacing with natural language.",
      "highlights": [
        "Pro TTS targets quality on complex prompts; Flash TTS targets cost-efficient everyday applications",
        "Input token limit of 8,192 per the Gemini API model pages",
        "Enhanced in December 2025 with improved expressivity, pacing and dialogue"
      ],
      "source_title": "Release notes - Gemini API | Google AI Studio",
      "source_url": "https://ai.google.dev/gemini-api/docs/changelog",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/05#gemini-2-5-tts"
    },
    {
      "slug": "gemma-3n",
      "date": "2025-05-20",
      "model": "Gemma 3n",
      "lab": "Google",
      "lab_slug": "google",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32768,
      "parameters": "E2B, E4B (5B and 8B raw)",
      "model_license": "Gemma Terms of Use",
      "available_at_launch": [
        "Open weights",
        "API",
        "Cloud platforms"
      ],
      "summary": "Gemma 3n is a mobile-first open model that accepts text, image, audio and video input on devices with limited memory. It was previewed in May 2025 and fully released on June 26, 2025.",
      "highlights": [
        "Per-Layer Embeddings let E2B and E4B run in about 2GB and 3GB of accelerator memory",
        "MatFormer architecture nests an E2B sub-model inside E4B",
        "Text support for 140 languages and multimodal understanding in 35",
        "Architecture shared with the next generation of Gemini Nano"
      ],
      "source_title": "Announcing Gemma 3n preview: powerful, efficient, mobile-first AI",
      "source_url": "https://developers.googleblog.com/en/introducing-gemma-3n/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/05#gemma-3n"
    },
    {
      "slug": "imagen-4",
      "date": "2025-05-20",
      "model": "Imagen 4",
      "lab": "Google",
      "lab_slug": "google",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API",
        "Cloud platforms"
      ],
      "summary": "Imagen 4 is Google DeepMind's text-to-image model announced at I/O 2025, with output up to 2K resolution and much better spelling and typography. It later shipped in Standard, Ultra and Fast variants.",
      "highlights": [
        "Launched in the Gemini app, Whisk, Vertex AI and Workspace apps such as Slides, Vids and Docs",
        "Imagen 4 and Imagen 4 Ultra entered preview in the Gemini API on June 24, 2025",
        "Imagen 4 Standard, Ultra and Fast generally available in the Gemini API and Vertex AI from August 14, 2025"
      ],
      "source_title": "Fuel your creativity with new generative media models and tools",
      "source_url": "https://blog.google/technology/ai/generative-media-models-io-2025/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/05#imagen-4"
    },
    {
      "slug": "medgemma",
      "date": "2025-05-20",
      "model": "MedGemma",
      "lab": "Google",
      "lab_slug": "google",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 131072,
      "parameters": "4B (multimodal), 27B (text)",
      "model_license": "Health AI Developer Foundations Terms of Use",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "MedGemma is a collection of Gemma 3-based open models for medical text and image comprehension, released through Google's Health AI Developer Foundations. The initial release included a 4B multimodal model and a 27B text-only model.",
      "highlights": [
        "4B model ships in pre-trained and instruction-tuned versions",
        "Supports at least 128K tokens of input and up to 8,192 output tokens",
        "Gated on Hugging Face under the Health AI Developer Foundations terms",
        "Intended as a starting point that requires validation before clinical use"
      ],
      "source_title": "google/medgemma-4b-it · Hugging Face",
      "source_url": "https://huggingface.co/google/medgemma-4b-it",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/05#medgemma"
    },
    {
      "slug": "veo-3",
      "date": "2025-05-20",
      "model": "Veo 3",
      "lab": "Google",
      "lab_slug": "google",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video",
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API",
        "Cloud platforms"
      ],
      "summary": "Veo 3 is Google DeepMind's video generation model announced at I/O 2025 and the first Veo model to generate audio with video, including dialogue and background sound. It launched in the Gemini app, Flow and Vertex AI.",
      "highlights": [
        "Initially for U.S. Google AI Ultra subscribers in the Gemini app and Flow, and enterprise users on Vertex AI",
        "Veo 3 Fast entered preview in the Gemini API on July 31, 2025",
        "Veo 3 and Veo 3 Fast generally available on Vertex AI from July 29, 2025 and in the Gemini API from September 9, 2025"
      ],
      "source_title": "Fuel your creativity with new generative media models and tools",
      "source_url": "https://blog.google/technology/ai/generative-media-models-io-2025/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/05#veo-3"
    },
    {
      "slug": "codex-1",
      "date": "2025-05-16",
      "model": "codex-1 and codex-mini-latest",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "coding",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "Enterprise",
        "API"
      ],
      "summary": "codex-1 is a version of OpenAI o3 optimized for software engineering that powers the Codex cloud agent research preview. OpenAI also released codex-mini-latest, a version of o4-mini built for the Codex CLI.",
      "highlights": [
        "Codex rolled out first to ChatGPT Pro, Enterprise, and Business users; Plus followed June 3, 2025",
        "codex-mini-latest became the default Codex CLI model and an API model",
        "codex-1 was trained with reinforcement learning on real-world coding tasks",
        "codex-mini-latest has a 200,000-token context window per the API model page"
      ],
      "source_title": "Introducing Codex | OpenAI",
      "source_url": "https://openai.com/index/introducing-codex/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/05#codex-1"
    },
    {
      "slug": "stable-audio-open-small",
      "date": "2025-05-14",
      "model": "Stable Audio Open Small",
      "lab": "Stability AI",
      "lab_slug": "stability-ai",
      "type": "music",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": "341M",
      "model_license": "Stability AI Community License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "341M-parameter text-to-audio model developed with Arm and optimized to run entirely on Arm CPUs, generating up to 11 seconds of audio on a smartphone in under 8 seconds.",
      "highlights": [
        "341M parameters versus 1.1B for Stable Audio Open",
        "Optimized with Arm KleidiAI",
        "Free for commercial use under the Stability AI Community License"
      ],
      "source_title": "Stability AI and Arm Collaborate to Release Stable Audio Open Small, Enabling Real-World Deployment for On-Device Audio Generation",
      "source_url": "https://stability.ai/news-updates/stability-ai-and-arm-release-stable-audio-open-small-enabling-real-world-deployment-for-on-device-audio-control",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/05#stable-audio-open-small"
    },
    {
      "slug": "wan2-1-vace",
      "date": "2025-05-14",
      "model": "Wan2.1-VACE",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": "1.3B, 14B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Wan2.1-VACE is an all-in-one model for video creation and editing in the Wan2.1 family, released with inference code, weights, and a technical report.",
      "highlights": [
        "Released in 1.3B and 14B sizes",
        "Handles multiple video creation and editing tasks in one model",
        "Released under the Apache 2.0 license"
      ],
      "source_title": "Wan2.1 (GitHub)",
      "source_url": "https://github.com/Wan-Video/Wan2.1",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/05#wan2-1-vace"
    },
    {
      "slug": "seed1-5-vl",
      "date": "2025-05-13",
      "model": "Seed1.5-VL",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "MoE LLM with 20B active, plus 532M vision encoder",
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Seed1.5-VL is ByteDance Seed's vision-language model, pairing a 532M-parameter SeedViT encoder with a mixture-of-experts language model of 20B active parameters for image, video, GUI, and game understanding and reasoning.",
      "highlights": [
        "Pretrained on more than 3T tokens of multimodal data",
        "API on Volcano Engine as Doubao-1.5-thinking-vision-pro",
        "Technical report and sample code published"
      ],
      "source_title": "First Release of Seed VLM Tech Report: Comprehensive Solutions for Image, Video, GUI, and Game",
      "source_url": "https://seed.bytedance.com/en/blog/first-release-of-seed-vlm-tech-report-comprehensive-solutions-for-image-video-gui-and-game",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/05#seed1-5-vl"
    },
    {
      "slug": "step1x-3d",
      "date": "2025-05-13",
      "model": "Step1X-3D",
      "lab": "StepFun",
      "lab_slug": "stepfun",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "image"
      ],
      "output_modalities": [
        "3d"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Step1X-3D is an open framework for generating textured 3D assets, pairing a geometry generation model with a texture synthesis model, released with training code and a curated 3D dataset.",
      "highlights": [
        "1.3B-parameter geometry model and 3.5B-parameter texture model",
        "Training code and 800K curated 3D asset IDs released",
        "Weights on Hugging Face under Apache-2.0"
      ],
      "source_title": "GitHub - stepfun-ai/Step1X-3D",
      "source_url": "https://github.com/stepfun-ai/Step1X-3D",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/05#step1x-3d"
    },
    {
      "slug": "seed1-5-embedding",
      "date": "2025-05-12",
      "model": "Seed1.5-Embedding",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "embedding",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Seed1.5-Embedding is a text embedding model further trained from Seed1.5 (Doubao-1.5-pro), optimized for general retrieval and reasoning-intensive search, using a mixture-of-experts backbone with few active parameters and multiple vector dimensions.",
      "highlights": [
        "Two-stage contrastive training on curated data",
        "Supports multiple embedding dimensions via MRL training",
        "API available through Volcano Ark as doubao-embedding-large-text"
      ],
      "source_title": "ByteDance's Seed1.5-Embedding Model Achieves SOTA in Retrieval: Training Details Unveiled",
      "source_url": "https://seed.bytedance.com/en/blog/bytedance-s-seed1-5-embedding-model-achieves-sota-in-retrieval-training-details-unveiled",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/05#seed1-5-embedding"
    },
    {
      "slug": "hunyuancustom",
      "date": "2025-05-08",
      "model": "HunyuanCustom",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Tencent Hunyuan Community License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "HunyuanCustom is an open multimodal-driven architecture for customized video generation built on HunyuanVideo, starting with single-subject customization and later adding audio-driven and video-driven customization.",
      "highlights": [
        "Audio-driven and video-driven weights released June 6, 2025",
        "Runs on a single GPU with 8GB VRAM through community tooling"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/HunyuanCustom: HunyuanCustom: A Multimodal-Driven Architecture for Customized Video Generation",
      "source_url": "https://github.com/Tencent-Hunyuan/HunyuanCustom",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/05#hunyuancustom"
    },
    {
      "slug": "seed-coder",
      "date": "2025-05-08",
      "model": "Seed-Coder",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32768,
      "parameters": "8B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Seed-Coder is an open-weights family of 8B code models with base, instruct, and reasoning variants, pretrained on 6T tokens of code data curated largely by an LLM-driven filtering pipeline.",
      "highlights": [
        "Base and Instruct support 32K context; Reasoning supports 64K",
        "Released under the MIT License",
        "A BF16 version of the reasoning model was added later"
      ],
      "source_title": "GitHub - ByteDance-Seed/Seed-Coder: Seed-Coder is a family of lightweight open-source code LLMs comprising base, instruct and reasoning models, developed by ByteDance Seed.",
      "source_url": "https://github.com/ByteDance-Seed/Seed-Coder",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/05#seed-coder"
    },
    {
      "slug": "mistral-medium-3",
      "date": "2025-05-07",
      "model": "Mistral Medium 3",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Enterprise"
      ],
      "summary": "Enterprise-focused multimodal model that Mistral positioned as balancing strong performance with lower cost and simpler deployment, with particular strength in coding and multimodal understanding.",
      "highlights": [
        "128K-token context window",
        "Deployable on any cloud or self-hosted on four or more GPUs",
        "API model ID mistral-medium-2505"
      ],
      "source_title": "Medium is the new large. | Mistral AI",
      "source_url": "https://mistral.ai/news/mistral-medium-3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/05#mistral-medium-3"
    },
    {
      "slug": "gemini-2-5-pro-io-edition",
      "date": "2025-05-06",
      "model": "Gemini 2.5 Pro Preview (I/O edition)",
      "lab": "Google",
      "lab_slug": "google",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms",
        "Consumer app"
      ],
      "summary": "Gemini 2.5 Pro Preview (I/O edition) was an updated 2.5 Pro released ahead of Google I/O 2025 with significantly improved coding capabilities, especially for building interactive web apps.",
      "highlights": [
        "Model ID gemini-2.5-pro-preview-05-06; the 03-25 preview automatically pointed to the new version",
        "Available in the Gemini API via Google AI Studio and Vertex AI and in the Gemini app",
        "Further updated as gemini-2.5-pro-preview-06-05 before general availability"
      ],
      "source_title": "Gemini 2.5 Pro update: Coding, web apps with Gemini",
      "source_url": "https://blog.google/products-and-platforms/products/gemini/gemini-2-5-pro-updates/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/05#gemini-2-5-pro-io-edition"
    },
    {
      "slug": "ltx-video-13b-0-9-7",
      "date": "2025-05-05",
      "model": "LTX-Video 13B 0.9.7",
      "lab": "Lightricks",
      "lab_slug": "lightricks",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": "13B",
      "model_license": "LTXV Open Weights License 0.X",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "13B-parameter LTX-Video model with improved prompt adherence and physical understanding and a multi-scale rendering pipeline, released with FP8 quantized weights and spatial and temporal upscalers.",
      "highlights": [
        "13B parameters",
        "Distilled 13B model followed on May 14, 2025",
        "Licensed under the LTXV Open Weights License"
      ],
      "source_title": "Lightricks/LTX-Video",
      "source_url": "https://github.com/Lightricks/LTX-Video",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/05#ltx-video-13b-0-9-7"
    },
    {
      "slug": "granite-4-0-tiny-preview",
      "date": "2025-05-02",
      "model": "Granite 4.0 Tiny Preview",
      "lab": "IBM",
      "lab_slug": "ibm",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "MoE: 7B total",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Early preview of the Granite 4.0 architecture, a 7B fine-grained hybrid mixture-of-experts instruct model released ahead of the full Granite 4.0 family.",
      "highlights": [
        "Fine-grained hybrid MoE architecture",
        "Fine-tuned from Granite-4.0-Tiny-Base-Preview",
        "Released under Apache 2.0"
      ],
      "source_title": "ibm-granite/granite-4.0-tiny-preview · Hugging Face",
      "source_url": "https://huggingface.co/ibm-granite/granite-4.0-tiny-preview",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/05#granite-4-0-tiny-preview"
    },
    {
      "slug": "xgen-small",
      "date": "2025-05-02",
      "model": "xGen-small",
      "lab": "Salesforce",
      "lab_slug": "salesforce",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "4B and 9B",
      "model_license": "CC-BY-NC-4.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Compact enterprise-focused language models trained from scratch in 4B and 9B sizes, combining domain-focused data curation, length extension, instruction tuning and reinforcement learning for long-context performance at low cost.",
      "highlights": [
        "Trained on a budget of eight trillion tokens",
        "Base and instruct variants supported 128k-token context",
        "Released for research use under CC BY-NC 4.0"
      ],
      "source_title": "xGen-small: Enterprise-ready Small Language Models",
      "source_url": "https://www.salesforce.com/blog/xgen-small-enterprise-ready-small-language-models/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/05#xgen-small"
    },
    {
      "slug": "parakeet-tdt-0-6b-v2",
      "date": "2025-05-01",
      "model": "Parakeet TDT 0.6B v2",
      "lab": "NVIDIA",
      "lab_slug": "nvidia",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "600M",
      "model_license": "CC-BY-4.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "600-million-parameter English automatic speech recognition model from NVIDIA with punctuation, capitalization and accurate timestamp prediction, released openly under CC BY 4.0.",
      "highlights": [
        "Supports English transcription",
        "Predicts word-level timestamps accurately",
        "Succeeded by the multilingual v3 in August 2025"
      ],
      "source_title": "nvidia/parakeet-tdt-0.6b-v2 · Hugging Face",
      "source_url": "https://huggingface.co/nvidia/parakeet-tdt-0.6b-v2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/05#parakeet-tdt-0-6b-v2"
    },
    {
      "slug": "suno-v4-5",
      "date": "2025-05-01",
      "model": "Suno v4.5",
      "lab": "Suno",
      "lab_slug": "suno",
      "type": "music",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "More expressive Suno model with more dynamic music, greater genre variety and accuracy, and richer vocals, released with tools to combine Personas, Covers and Extend.",
      "highlights": [
        "Greater variety and accuracy in genres",
        "Richer vocals",
        "Longer Covers and combined features"
      ],
      "source_title": "Introducing v4.5",
      "source_url": "https://suno.com/blog/introducing-v4-5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/05#suno-v4-5"
    },
    {
      "slug": "amazon-nova-premier",
      "date": "2025-04-30",
      "model": "Amazon Nova Premier",
      "lab": "Amazon",
      "lab_slug": "amazon",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Amazon Nova Premier is the most capable model in the first Amazon Nova generation, built for complex tasks and to serve as a teacher for distilling smaller custom models, with a one-million-token context window.",
      "highlights": [
        "Accepts text, image and video input; audio is not supported",
        "One-million-token context window",
        "First announced as coming on December 3, 2024"
      ],
      "source_title": "Amazon Nova Premier: Our most capable model for complex tasks and teacher for model distillation",
      "source_url": "https://aws.amazon.com/blogs/aws/amazon-nova-premier-our-most-capable-model-for-complex-tasks-and-teacher-for-model-distillation/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#amazon-nova-premier"
    },
    {
      "slug": "deepseek-prover-v2",
      "date": "2025-04-30",
      "model": "DeepSeek-Prover-V2",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "7B, 671B",
      "model_license": "DeepSeek Model License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "DeepSeek-Prover-V2 is an open-weights model for formal theorem proving in Lean 4, trained with cold-start data from a recursive proof-search pipeline powered by DeepSeek-V3. It comes in 7B and 671B sizes.",
      "highlights": [
        "DeepSeek-Prover-V2-671B is trained on top of DeepSeek-V3-Base",
        "DeepSeek-Prover-V2-7B builds on Prover-V1.5-Base with context extended to 32K tokens",
        "Use is subject to the DeepSeek model license"
      ],
      "source_title": "DeepSeek-Prover-V2 (GitHub)",
      "source_url": "https://github.com/deepseek-ai/DeepSeek-Prover-V2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#deepseek-prover-v2"
    },
    {
      "slug": "runway-gen-4-references",
      "date": "2025-04-30",
      "model": "Gen-4 References (Gen-4 Image)",
      "lab": "Runway",
      "lab_slug": "runway",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Gen-4 image generation model that uses reference images to create consistent characters, locations and other elements, launched as Gen-4 References for paid plans.",
      "highlights": [
        "Extended to Free plan users on May 7, 2025",
        "Gen-4 Image API launched May 16, 2025",
        "Improved object consistency update on June 25, 2025"
      ],
      "source_title": "Product Updates & Changelog | Runway AI",
      "source_url": "https://runway.com/changelog",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#runway-gen-4-references"
    },
    {
      "slug": "mimo-7b",
      "date": "2025-04-30",
      "model": "MiMo-7B",
      "lab": "Xiaomi",
      "lab_slug": "xiaomi",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "7B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "MiMo-7B is Xiaomi's first open-source language model series, built for math and code reasoning and released as base, SFT, RL-Zero, and RL checkpoints with pre-training and post-training tailored to reasoning.",
      "highlights": [
        "Base model pre-trained on about 25 trillion tokens with multiple-token prediction",
        "MiMo-7B-RL-0530 update on May 30, 2025 scaled SFT data and extended the RL window to 48K",
        "Weights under the MIT license"
      ],
      "source_title": "XiaomiMiMo/MiMo",
      "source_url": "https://github.com/XiaomiMiMo/MiMo",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#mimo-7b"
    },
    {
      "slug": "phi-4-mini-reasoning",
      "date": "2025-04-30",
      "model": "Phi-4-mini-reasoning",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "3.8B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Phi-4-mini-reasoning is a 3.8B open Microsoft model for math reasoning, fine-tuned from Phi-4-mini with synthetic data generated by DeepSeek-R1, and aimed at edge, mobile and educational applications.",
      "highlights": [
        "3.8B parameters with a 128K token context length",
        "Fine-tuned on synthetic data generated by DeepSeek-R1",
        "ONNX version for Copilot+ PCs made available May 15, 2025"
      ],
      "source_title": "One year of Phi: Small language models making big leaps in AI",
      "source_url": "https://azure.microsoft.com/en-us/blog/one-year-of-phi-small-language-models-making-big-leaps-in-ai/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#phi-4-mini-reasoning"
    },
    {
      "slug": "phi-4-reasoning",
      "date": "2025-04-30",
      "model": "Phi-4-reasoning and Phi-4-reasoning-plus",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32000,
      "parameters": "14B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Phi-4-reasoning is a 14B open-weight reasoning model fine-tuned from Phi-4 on reasoning demonstrations, released with Phi-4-reasoning-plus, which adds reinforcement learning to use more inference-time compute.",
      "highlights": [
        "Both models have 14B parameters and a 32K token context length",
        "Phi-4-reasoning-plus is further trained with reinforcement learning",
        "Available on Azure AI Foundry and Hugging Face under the MIT license"
      ],
      "source_title": "One year of Phi: Small language models making big leaps in AI",
      "source_url": "https://azure.microsoft.com/en-us/blog/one-year-of-phi-small-language-models-making-big-leaps-in-ai/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#phi-4-reasoning"
    },
    {
      "slug": "llama-guard-4",
      "date": "2025-04-29",
      "model": "Llama Guard 4",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "12B",
      "model_license": "Llama 4 Community License",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Llama Guard 4 is a 12B natively multimodal safety classifier that moderates text and multi-image inputs and model responses, acting as a unified safeguard across modalities.",
      "highlights": [
        "Dense model pruned from the Llama 4 Scout pretrained model",
        "Classifies content against 14 MLCommons-based hazard categories",
        "Also offered on the Llama API, then in limited preview"
      ],
      "source_title": "Sharing new open source protection tools and advancements in AI privacy and security",
      "source_url": "https://ai.meta.com/blog/ai-defenders-program-llama-protection-tools/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#llama-guard-4"
    },
    {
      "slug": "llama-prompt-guard-2",
      "date": "2025-04-29",
      "model": "Llama Prompt Guard 2",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 512,
      "parameters": "86M, 22M",
      "model_license": "Llama 4 Community License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Llama Prompt Guard 2 updates Meta's prompt attack classifier with an improved 86M model and a smaller 22M model for detecting prompt injection and jailbreak attempts.",
      "highlights": [
        "The 22M model cuts latency and compute costs by up to 75% versus the 86M model",
        "86M is based on mDeBERTa-base; 22M on DeBERTa-xsmall",
        "Both support a 512-token context window"
      ],
      "source_title": "Sharing new open source protection tools and advancements in AI privacy and security",
      "source_url": "https://ai.meta.com/blog/ai-defenders-program-llama-protection-tools/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#llama-prompt-guard-2"
    },
    {
      "slug": "qwen3",
      "date": "2025-04-29",
      "model": "Qwen3",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32768,
      "parameters": "MoE: 235B total, 22B active (flagship); 0.6B to 32B dense",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "Qwen3 is Alibaba's open-weights model family of two mixture-of-experts models and six dense models from 0.6B to 32B parameters. The models offer hybrid thinking and non-thinking modes with user control over how much reasoning a task receives.",
      "highlights": [
        "Pre-trained on about 36T tokens covering 119 languages and dialects",
        "Flagship Qwen3-235B-A22B supports 32,768 tokens natively and 131,072 with YaRN",
        "Weights under the Apache 2.0 license"
      ],
      "source_title": "Qwen3: Think Deeper, Act Faster",
      "source_url": "https://qwenlm.github.io/blog/qwen3/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#qwen3"
    },
    {
      "slug": "ernie-4-5-turbo",
      "date": "2025-04-25",
      "model": "ERNIE 4.5 Turbo",
      "lab": "Baidu",
      "lab_slug": "baidu",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "ERNIE 4.5 Turbo is a faster multimodal model derived from ERNIE 4.5, launched at Baidu Create 2025 with improvements in hallucination reduction, logical reasoning, and coding.",
      "highlights": [
        "Launched at Baidu Create 2025 in Wuhan",
        "Enhanced multimodal capabilities, launched alongside ERNIE X1 Turbo",
        "Serves as the base for the ERNIE X1 Turbo reasoning model"
      ],
      "source_title": "文心4.5Turbo、X1Turbo和多款AI应用发布！李彦宏：应用创造未来",
      "source_url": "https://home.baidu.com/home/index/news_detail/id/18023",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#ernie-4-5-turbo"
    },
    {
      "slug": "ernie-x1-turbo",
      "date": "2025-04-25",
      "model": "ERNIE X1 Turbo",
      "lab": "Baidu",
      "lab_slug": "baidu",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "ERNIE X1 Turbo is a deep-thinking model built on ERNIE 4.5 Turbo, launched at Baidu Create 2025 with a more advanced chain of thought and stronger question answering, writing, reasoning, tool use, and multimodal abilities.",
      "highlights": [
        "Launched with ERNIE 4.5 Turbo at Create 2025 in Wuhan",
        "Combines thinking with tool calls in a compound chain of thought and action",
        "Built on ERNIE 4.5 Turbo, which is derived from ERNIE 4.5"
      ],
      "source_title": "百度发布文心4.5 Turbo、X1 Turbo两大模型，百度王海峰详解背后技术",
      "source_url": "https://home.baidu.com/home/index/news_detail/id/18025",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#ernie-x1-turbo"
    },
    {
      "slug": "kimi-audio",
      "date": "2025-04-25",
      "model": "Kimi-Audio",
      "lab": "Moonshot AI",
      "lab_slug": "moonshot-ai",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": null,
      "parameters": "7B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Kimi-Audio is an open-source audio foundation model for audio understanding, generation, and conversation in one framework. It handles tasks such as speech recognition, audio question answering, captioning, emotion recognition, and end-to-end speech conversation.",
      "highlights": [
        "Pre-trained on more than 13 million hours of audio and text data",
        "Hybrid input of continuous acoustic and discrete semantic tokens",
        "Kimi-Audio-7B-Instruct released April 25, 2025; base Kimi-Audio-7B weights followed April 27"
      ],
      "source_title": "Kimi-Audio (GitHub)",
      "source_url": "https://github.com/MoonshotAI/Kimi-Audio",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#kimi-audio"
    },
    {
      "slug": "step1x-edit",
      "date": "2025-04-25",
      "model": "Step1X-Edit",
      "lab": "StepFun",
      "lab_slug": "stepfun",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Step1X-Edit is StepFun's open-source general image editing model that follows natural-language editing instructions, released with inference code, weights, a technical report, and the GEdit-Bench evaluation benchmark.",
      "highlights": [
        "Released with the GEdit-Bench image editing benchmark",
        "Updated v1.1 weights followed on July 9, 2025",
        "Weights on Hugging Face and ModelScope under Apache-2.0"
      ],
      "source_title": "GitHub - stepfun-ai/Step1X-Edit",
      "source_url": "https://github.com/stepfun-ai/Step1X-Edit",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#step1x-edit"
    },
    {
      "slug": "lyria-2",
      "date": "2025-04-24",
      "model": "Lyria 2",
      "lab": "Google",
      "lab_slug": "google",
      "type": "music",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "Cloud platforms"
      ],
      "summary": "Lyria 2 is Google DeepMind's music generation model for high-fidelity, professional-grade audio, introduced with an expanded Music AI Sandbox. It later became generally available to enterprises on Vertex AI and to creators through YouTube Shorts.",
      "highlights": [
        "All Lyria 2 output is watermarked with SynthID",
        "Music AI Sandbox access expanded to more U.S. musicians, producers and songwriters",
        "Generally available on Vertex AI from May 20, 2025"
      ],
      "source_title": "Music AI Sandbox, now with new features and broader access",
      "source_url": "https://deepmind.google/discover/blog/music-ai-sandbox-now-with-new-features-and-broader-access/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#lyria-2"
    },
    {
      "slug": "lyria-realtime",
      "date": "2025-04-24",
      "model": "Lyria RealTime",
      "lab": "Google",
      "lab_slug": "google",
      "type": "music",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Lyria RealTime is Google DeepMind's interactive music generation model that lets users create, perform and steer a continuous music stream in real time, mixing genres and blending styles as it plays. It powers MusicFX DJ.",
      "highlights": [
        "Output is watermarked with SynthID",
        "Released in the Gemini API and AI Studio as lyria-realtime-exp on May 20, 2025",
        "Its open-weights counterpart is Magenta RealTime"
      ],
      "source_title": "Music AI Sandbox, now with new features and broader access",
      "source_url": "https://deepmind.google/discover/blog/music-ai-sandbox-now-with-new-features-and-broader-access/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#lyria-realtime"
    },
    {
      "slug": "glm-asr",
      "date": "2025-04-23",
      "model": "GLM-ASR",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "GLM-ASR is Z.ai's speech recognition model on the BigModel API, which uses context to transcribe audio into idiomatic, readable text. Z.ai says it performs better in noisy environments than its earlier models.",
      "highlights": [
        "Supports Chinese, English, and several regional Chinese dialects",
        "Not disrupted by non-speech noise",
        "Available through the BigModel API platform"
      ],
      "source_title": "Model and product release records (BigModel docs)",
      "source_url": "https://docs.bigmodel.cn/cn/update/new-releases",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#glm-asr"
    },
    {
      "slug": "gemma-3-qat",
      "date": "2025-04-18",
      "model": "Gemma 3 QAT",
      "lab": "Google",
      "lab_slug": "google",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 131072,
      "parameters": "1B, 4B, 12B, 27B",
      "model_license": "Gemma Terms of Use",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Gemma 3 QAT is a set of quantization-aware trained int4 Gemma 3 checkpoints that cut memory needs sharply while preserving quality. The 27B model's weights shrink from 54GB in BF16 to 14.1GB, fitting a single consumer GPU.",
      "highlights": [
        "Covers the 27B, 12B, 4B and 1B Gemma 3 sizes",
        "27B fits on a single 24GB card such as the NVIDIA RTX 3090",
        "Official int4 and Q4_0 builds on Hugging Face and Kaggle; supported in Ollama, LM Studio, MLX and llama.cpp"
      ],
      "source_title": "Gemma 3 QAT Models: Bringing state-of-the-Art AI to consumer GPUs",
      "source_url": "https://developers.googleblog.com/en/gemma-3-quantized-aware-trained-state-of-the-art-ai-to-consumer-gpus/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#gemma-3-qat"
    },
    {
      "slug": "gemini-2-5-flash",
      "date": "2025-04-17",
      "model": "Gemini 2.5 Flash",
      "lab": "Google",
      "lab_slug": "google",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1048576,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms",
        "Consumer app"
      ],
      "summary": "Gemini 2.5 Flash is Google's price-performance Gemini 2.5 model, released in preview with a major upgrade in reasoning over 2.0 Flash and a thinking budget that lets developers control how much the model thinks.",
      "highlights": [
        "Preview model gemini-2.5-flash-preview-04-17 in the Gemini API via Google AI Studio and Vertex AI",
        "An updated preview (gemini-2.5-flash-preview-05-20) shipped at Google I/O 2025",
        "Stable gemini-2.5-flash became generally available on June 17, 2025",
        "Also available to everyone in the Gemini app"
      ],
      "source_title": "Start building with Gemini 2.5 Flash",
      "source_url": "https://developers.googleblog.com/start-building-with-gemini-25-flash/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#gemini-2-5-flash"
    },
    {
      "slug": "mai-ds-r1",
      "date": "2025-04-17",
      "model": "MAI-DS-R1",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "MAI-DS-R1 is an open-weights version of DeepSeek R1 post-trained by the Microsoft AI team to respond to more previously blocked topics and reduce risk while keeping the original model's reasoning capabilities.",
      "highlights": [
        "Post-trained from DeepSeek-R1",
        "Released on Azure AI Foundry and Hugging Face",
        "Retired from Microsoft Foundry on February 27, 2026"
      ],
      "source_title": "Introducing MAI-DS-R1",
      "source_url": "https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/introducing-mai-ds-r1/4405076",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#mai-ds-r1"
    },
    {
      "slug": "granite-3-3",
      "date": "2025-04-16",
      "model": "Granite 3.3",
      "lab": "IBM",
      "lab_slug": "ibm",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "2B and 8B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Update of the Granite 2B and 8B models with improved reasoning that can be toggled, fill-in-the-middle support and a 128K-token context for the 8B model.",
      "highlights": [
        "Released alongside Granite Speech 3.3 8B, an audio-in, text-out speech model",
        "Base models supersede their Granite 3.1 predecessors",
        "Released under Apache 2.0"
      ],
      "source_title": "IBM Granite 3.3: Speech recognition, refined reasoning, and RAG LoRAs",
      "source_url": "https://www.ibm.com/new/announcements/ibm-granite-3-3-speech-recognition-refined-reasoning-rag-loras",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#granite-3-3"
    },
    {
      "slug": "o3",
      "date": "2025-04-16",
      "model": "OpenAI o3 and o4-mini",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 200000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "o3 and o4-mini are o-series reasoning models and the first that can agentically use and combine every ChatGPT tool, including web search, Python, file analysis, visual reasoning, and image generation.",
      "highlights": [
        "API model IDs o3 and o4-mini (snapshots dated 2025-04-16)",
        "Replaced o1, o3-mini, and o3-mini-high for ChatGPT Plus, Pro, and Team users",
        "ChatGPT also offered o4-mini-high; free users could try o4-mini via 'Think'",
        "200,000-token context window per the API model pages"
      ],
      "source_title": "Introducing OpenAI o3 and o4-mini | OpenAI",
      "source_url": "https://openai.com/index/introducing-o3-and-o4-mini/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#o3"
    },
    {
      "slug": "seedream-3-0",
      "date": "2025-04-16",
      "model": "Seedream 3.0",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Seedream 3.0 is ByteDance Seed's bilingual Chinese-English text-to-image model, adding native 2K output, faster generation, and more accurate small-text rendering and layout compared with Seedream 2.0.",
      "highlights": [
        "Native 2K output across aspect ratios without post-processing",
        "Launched in April 2025 and fully available in Doubao and Jimeng",
        "Volcano Engine API model ID doubao-seedream-3.0-t2i-250415"
      ],
      "source_title": "Seedream 3.0 Text-to-Image Model Technical Report Released",
      "source_url": "https://seed.bytedance.com/en/blog/seedream-3-0-text-to-image-model-technical-report-released",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#seedream-3-0"
    },
    {
      "slug": "ui-tars-1-5",
      "date": "2025-04-16",
      "model": "UI-TARS-1.5",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "7B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "UI-TARS-1.5 is a multimodal GUI agent built on a vision-language model and improved with reinforcement learning so it reasons before acting in computer-use and game tasks; ByteDance Seed open-sourced a 7B version.",
      "highlights": [
        "UI-TARS-1.5-7B weights released on Hugging Face under Apache-2.0",
        "Adds game scenarios supporting up to 1,000 interactions",
        "The open 7B model targets general computer use rather than games"
      ],
      "source_title": "GitHub - bytedance/UI-TARS: Pioneering Automated GUI Interaction with Native Agents",
      "source_url": "https://github.com/bytedance/UI-TARS",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#ui-tars-1-5"
    },
    {
      "slug": "cohere-embed-4",
      "date": "2025-04-15",
      "model": "Embed 4",
      "lab": "Cohere",
      "lab_slug": "cohere",
      "type": "embedding",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "embedding"
      ],
      "context_window": 128000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Multimodal embedding model for searching complex documents such as PDF reports and slides that mix text, images, tables, graphs and code.",
      "highlights": [
        "Embeds documents up to 128K tokens, about 200 pages",
        "Supports 100+ languages and cross-lingual search",
        "Matryoshka embeddings for flexible storage"
      ],
      "source_title": "Introducing Embed 4: Multimodal search for business | Cohere Blog",
      "source_url": "https://cohere.com/blog/embed-4",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#cohere-embed-4"
    },
    {
      "slug": "kling-2-0",
      "date": "2025-04-15",
      "model": "Kling 2.0",
      "lab": "Kuaishou",
      "lab_slug": "kuaishou",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Kling 2.0 Master is the second-generation Kling video generation model, launched globally with improvements in motion quality, semantic responsiveness, and visual aesthetics, alongside the Kolors 2.0 image model.",
      "highlights": [
        "Launched at Kling's From Vision to Screen event in Beijing",
        "Released together with Kolors 2.0 and new multimodal editing features",
        "Kuaishou reported more than 20 Kling iterations since June 2024"
      ],
      "source_title": "Kling AI Advances to the 2.0 Era, Empowering Everyone to Tell Great Stories with AI",
      "source_url": "https://ir.kuaishou.com/news-releases/news-release-details/kling-ai-advances-20-era-empowering-everyone-tell-great-stories",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#kling-2-0"
    },
    {
      "slug": "kolors-2-0",
      "date": "2025-04-15",
      "model": "Kolors 2.0",
      "lab": "Kuaishou",
      "lab_slug": "kuaishou",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Kolors 2.0 is Kuaishou's second-generation image generation model, launched on Kling AI with the Kling 2.0 video model and improved prompt adherence, cinematic quality, and artistic style expression.",
      "highlights": [
        "Launched on Kling AI alongside Kling 2.0 Master",
        "Improves prompt adherence, cinematic quality, and artistic styles"
      ],
      "source_title": "Kling AI Advances to the 2.0 Era, Empowering Everyone to Tell Great Stories with AI",
      "source_url": "https://ir.kuaishou.com/news-releases/news-release-details/kling-ai-advances-20-era-empowering-everyone-tell-great-stories",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#kolors-2-0"
    },
    {
      "slug": "glm-4-0414",
      "date": "2025-04-14",
      "model": "GLM-4-0414 and GLM-Z1",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "9B, 32B",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "GLM-4-0414 is Z.ai's open-weights model series at 9B and 32B, released with the GLM-Z1 deep reasoning models and GLM-Z1-Rumination, which thinks longer on open-ended research tasks. GLM-4-32B-Base-0414 was pre-trained on 15T tokens.",
      "highlights": [
        "GLM-Z1-32B-0414 adds cold-start and extended reinforcement learning for math, code, and logic",
        "API launched GLM-4-Air-250414, GLM-4-Flash-250414, and GLM-Z1-Air, AirX, and Flash",
        "Weights under the MIT license"
      ],
      "source_title": "GLM-4 (GitHub)",
      "source_url": "https://github.com/zai-org/GLM-4",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#glm-4-0414"
    },
    {
      "slug": "gpt-4-1",
      "date": "2025-04-14",
      "model": "GPT-4.1, GPT-4.1 mini, and GPT-4.1 nano",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1047576,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "GPT-4.1, GPT-4.1 mini, and GPT-4.1 nano are API models that improve on GPT-4o in coding and instruction following, with context windows up to 1 million tokens and a June 2024 knowledge cutoff.",
      "highlights": [
        "API model IDs gpt-4.1, gpt-4.1-mini, and gpt-4.1-nano",
        "1,047,576-token context window for all three, per the API model pages",
        "GPT-4.1 nano is OpenAI's first nano model",
        "GPT-4.1 came to ChatGPT for all paid users on May 14, 2025"
      ],
      "source_title": "Introducing GPT-4.1 in the API | OpenAI",
      "source_url": "https://openai.com/index/gpt-4-1/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#gpt-4-1"
    },
    {
      "slug": "nemotron-h",
      "date": "2025-04-14",
      "model": "Nemotron-H",
      "lab": "NVIDIA",
      "lab_slug": "nvidia",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "NVIDIA Internal Scientific Research and Development Model License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Family of hybrid base language models from NVIDIA that replace most attention layers with Mamba-2 layers to speed up inference, released under a research license.",
      "highlights": [
        "The 8B base model uses only four attention layers",
        "Described in the Nemotron-H technical report",
        "Base completion models intended for customization"
      ],
      "source_title": "nvidia/Nemotron-H-8B-Base-8K · Hugging Face",
      "source_url": "https://huggingface.co/nvidia/Nemotron-H-8B-Base-8K",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#nemotron-h"
    },
    {
      "slug": "seed-thinking-v1-5",
      "date": "2025-04-14",
      "model": "Seed-Thinking-v1.5",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "MoE: 200B total, 20B active",
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Seed-Thinking-v1.5 is ByteDance Seed's reasoning model trained with large-scale reinforcement learning for math, coding, science, and creative writing; Volcano Engine serves it as Doubao-1.5-thinking-pro.",
      "highlights": [
        "Mixture-of-experts with 200B total and 20B active parameters",
        "API access on Volcano Engine opened April 17, 2025",
        "Technical report published on GitHub"
      ],
      "source_title": "ByteDance's Latest Thinking Model, Seed-Thinking-v1.5 Technical Details Disclosed",
      "source_url": "https://seed.bytedance.com/en/blog/bytedance-s-latest-thinking-model-seed-thinking-v1-5-technical-details-disclosed",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#seed-thinking-v1-5"
    },
    {
      "slug": "kimi-vl",
      "date": "2025-04-10",
      "model": "Kimi-VL",
      "lab": "Moonshot AI",
      "lab_slug": "moonshot-ai",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "MoE: 16B total, 3B active",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Kimi-VL is an efficient open-weights mixture-of-experts vision-language model for multimodal reasoning, long-context understanding, and agent tasks. It was released as Kimi-VL-A3B-Instruct and Kimi-VL-A3B-Thinking.",
      "highlights": [
        "Activates about 2.8B parameters in its language decoder",
        "128K extended context window",
        "Native-resolution MoonViT vision encoder for high-resolution inputs"
      ],
      "source_title": "moonshotai/Kimi-VL-A3B-Instruct (Hugging Face)",
      "source_url": "https://huggingface.co/moonshotai/Kimi-VL-A3B-Instruct",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#kimi-vl"
    },
    {
      "slug": "amazon-nova-sonic",
      "date": "2025-04-08",
      "model": "Amazon Nova Sonic",
      "lab": "Amazon",
      "lab_slug": "amazon",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": 300000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Amazon Nova Sonic is Amazon's first speech-to-speech foundation model, unifying speech understanding and generation in one model so developers can build real-time voice conversations without separate recognition and text-to-speech models.",
      "highlights": [
        "Launched with American and British English, adding Spanish in June 2025 and French, Italian and German in July 2025",
        "300K token context window",
        "Launched in Amazon Bedrock in US East (N. Virginia)"
      ],
      "source_title": "Introducing Amazon Nova Sonic: Human-like voice conversations for generative AI applications",
      "source_url": "https://aws.amazon.com/blogs/aws/introducing-amazon-nova-sonic-human-like-voice-conversations-for-generative-ai-applications/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#amazon-nova-sonic"
    },
    {
      "slug": "amazon-nova-reel-1-1",
      "date": "2025-04-07",
      "model": "Amazon Nova Reel 1.1",
      "lab": "Amazon",
      "lab_slug": "amazon",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Amazon Nova Reel 1.1 updates Amazon's video generation model to produce multi-shot videos of up to two minutes from 6-second shots, with style consistency across shots and improved single-shot quality and latency.",
      "highlights": [
        "Generates videos up to 2 minutes long",
        "Automated mode from one prompt or manual storyboard mode with up to 20 shots",
        "Available in Amazon Bedrock in US East (N. Virginia)"
      ],
      "source_title": "Amazon Nova Reel 1.1: Featuring up to 2-minutes multi-shot videos",
      "source_url": "https://aws.amazon.com/blogs/aws/amazon-nova-reel-1-1-featuring-up-to-2-minutes-multi-shot-videos/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#amazon-nova-reel-1-1"
    },
    {
      "slug": "runway-gen-4-turbo",
      "date": "2025-04-07",
      "model": "Gen-4 Turbo",
      "lab": "Runway",
      "lab_slug": "runway",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Faster and more efficient version of Gen-4, described by Runway as its fastest and most efficient video model at launch and made available to all users.",
      "highlights": [
        "Available to all plans at launch",
        "Turbo variant of Gen-4"
      ],
      "source_title": "Product Updates & Changelog | Runway AI",
      "source_url": "https://runway.com/changelog",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#runway-gen-4-turbo"
    },
    {
      "slug": "llama-3-1-nemotron-ultra-253b",
      "date": "2025-04-07",
      "model": "Llama-3.1-Nemotron-Ultra-253B-v1",
      "lab": "NVIDIA",
      "lab_slug": "nvidia",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "253B",
      "model_license": "NVIDIA Open Model License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Largest Llama Nemotron reasoning model, derived from Meta Llama-3.1-405B-Instruct with neural architecture search to cut memory use while keeping accuracy.",
      "highlights": [
        "Post-trained for reasoning, chat preferences, RAG and tool calling",
        "Trained between November 2024 and April 2025",
        "Part of the Llama Nemotron collection"
      ],
      "source_title": "nvidia/Llama-3_1-Nemotron-Ultra-253B-v1 · Hugging Face",
      "source_url": "https://huggingface.co/nvidia/Llama-3_1-Nemotron-Ultra-253B-v1",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#llama-3-1-nemotron-ultra-253b"
    },
    {
      "slug": "llama-4-maverick",
      "date": "2025-04-05",
      "model": "Llama 4 Maverick",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": "MoE: 400B total, 17B active, 128 experts",
      "model_license": "Llama 4 Community License",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms",
        "Consumer app"
      ],
      "summary": "Llama 4 Maverick is a natively multimodal mixture-of-experts model with 17B active parameters and 128 experts, released with Llama 4 Scout as the first Llama 4 models.",
      "highlights": [
        "400B total parameters; runs on a single NVIDIA H100 DGX host",
        "Model card lists a 1M-token context window and 12 supported languages",
        "Knowledge cutoff of August 2024",
        "Powers Meta AI in WhatsApp, Messenger, Instagram Direct and meta.ai"
      ],
      "source_title": "The Llama 4 herd: The beginning of a new era of natively multimodal AI innovation",
      "source_url": "https://ai.meta.com/blog/llama-4-multimodal-intelligence/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#llama-4-maverick"
    },
    {
      "slug": "llama-4-scout",
      "date": "2025-04-05",
      "model": "Llama 4 Scout",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 10000000,
      "parameters": "MoE: 109B total, 17B active, 16 experts",
      "model_license": "Llama 4 Community License",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "Llama 4 Scout is a natively multimodal mixture-of-experts model with 17B active parameters and 16 experts, notable for a 10 million token context window.",
      "highlights": [
        "109B total parameters; fits on a single NVIDIA H100 GPU with Int4 quantization",
        "Accepts multilingual text and image input",
        "Llama 4 Behemoth was previewed but not released because it was still training"
      ],
      "source_title": "The Llama 4 herd: The beginning of a new era of natively multimodal AI innovation",
      "source_url": "https://ai.meta.com/blog/llama-4-multimodal-intelligence/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#llama-4-scout"
    },
    {
      "slug": "leonardo-motion-2-0",
      "date": "2025-04-04",
      "model": "Motion 2.0",
      "lab": "Leonardo.Ai",
      "lab_slug": "leonardo-ai",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Leonardo.Ai's own video generation model, built into its AI Creation interface to make videos from a text prompt or from images, with boosted frame rates for smoother playback.",
      "highlights": [
        "Launched for paid users on web and mobile",
        "720p output added on May 6, 2025",
        "Offered on Replicate with Phoenix 1.0 from June 10, 2025"
      ],
      "source_title": "Leonardo.Ai on X: Double Drop at Leonardo.Ai. Motion 2.0 is here",
      "source_url": "https://x.com/LeonardoAi/status/1908089989210018111",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#leonardo-motion-2-0"
    },
    {
      "slug": "midjourney-v7",
      "date": "2025-04-03",
      "model": "Midjourney V7",
      "lab": "Midjourney",
      "lab_slug": "midjourney",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Seventh-generation image model with better text and image prompt handling, higher image quality and coherence, personalization on by default, and a new Draft Mode that renders at ten times the speed.",
      "highlights": [
        "Launched as an alpha with Turbo and Relax modes",
        "Became the default model on June 17, 2025",
        "Introduced Omni Reference and Draft Mode"
      ],
      "source_title": "Version - Midjourney",
      "source_url": "https://docs.midjourney.com/hc/en-us/articles/32199405667853-Version",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#midjourney-v7"
    },
    {
      "slug": "minimax-speech-02",
      "date": "2025-04-02",
      "model": "MiniMax Speech-02",
      "lab": "MiniMax",
      "lab_slug": "minimax",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Speech-02 is MiniMax's text-to-audio and voice cloning model series, offered as Speech-02-HD for high-fidelity voiceovers and audiobooks and Speech-02-Turbo for low-latency interactive use.",
      "highlights": [
        "Voice cloning from 10 seconds of recorded audio",
        "Library of more than 300 prebuilt voices",
        "Built on an autoregressive Transformer with a learnable speaker encoder, per MiniMax's May 2025 technical post"
      ],
      "source_title": "Speech-02-series : The Next Leap in Text-to-Audio and AI Voice Technology",
      "source_url": "https://www.minimax.io/news/speech-02-series",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/04#minimax-speech-02"
    },
    {
      "slug": "amazon-nova-act",
      "date": "2025-03-31",
      "model": "Amazon Nova Act",
      "lab": "Amazon",
      "lab_slug": "amazon",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Amazon Nova Act is an Amazon AGI model trained to perform actions within a web browser, released as a research preview SDK for building agents that complete multi-step browser tasks.",
      "highlights": [
        "Released as a research preview SDK through nova.amazon.com",
        "Agents can combine natural language commands with Python code and Playwright browser control",
        "Became an AWS service powered by a custom Nova 2 Lite model in December 2025"
      ],
      "source_title": "Introducing Amazon Nova Act",
      "source_url": "https://www.amazon.science/blog/nova-act",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#amazon-nova-act"
    },
    {
      "slug": "runway-gen-4",
      "date": "2025-03-31",
      "model": "Gen-4",
      "lab": "Runway",
      "lab_slug": "runway",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Video generation model built for world consistency, generating consistent characters, locations and objects across scenes from visual references and instructions without fine-tuning.",
      "highlights": [
        "Available to paid plans from April 1, 2025",
        "Regenerates elements from multiple perspectives within scenes",
        "Released with short films made to test narrative capabilities"
      ],
      "source_title": "Runway Gen-4: AI Video Generation with World Consistency",
      "source_url": "https://runway.com/research/introducing-runway-gen-4",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#runway-gen-4"
    },
    {
      "slug": "qvq-max",
      "date": "2025-03-28",
      "model": "QVQ-Max",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "QVQ-Max is the first official version of Qwen's visual reasoning model, following the exploratory QVQ-72B-Preview. It understands images and videos and reasons over them to solve problems.",
      "highlights": [
        "Accuracy improves as the maximum thinking length increases, per Qwen",
        "Covers math problems, everyday questions, code, and creative tasks",
        "Available to try on Qwen Chat"
      ],
      "source_title": "QVQ-Max: Think with Evidence",
      "source_url": "https://qwenlm.github.io/blog/qvq-max-preview/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#qvq-max"
    },
    {
      "slug": "qwen2-5-omni",
      "date": "2025-03-27",
      "model": "Qwen2.5-Omni",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": null,
      "parameters": "7B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Consumer app",
        "Open weights"
      ],
      "summary": "Qwen2.5-Omni is Qwen's end-to-end multimodal model that perceives text, images, audio, and video and responds with text and natural speech in a streaming manner. It uses a Thinker-Talker architecture.",
      "highlights": [
        "Released as Qwen2.5-Omni-7B; a 3B version followed on April 30, 2025",
        "Available on Qwen Chat",
        "Weights under the Apache 2.0 license"
      ],
      "source_title": "Qwen2.5 Omni: See, Hear, Talk, Write, Do It All!",
      "source_url": "https://qwenlm.github.io/blog/qwen2.5-omni/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#qwen2-5-omni"
    },
    {
      "slug": "ideogram-3-0",
      "date": "2025-03-26",
      "model": "Ideogram 3.0",
      "lab": "Ideogram",
      "lab_slug": "ideogram",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Ideogram image model focused on photorealism, legible text and creative designs, and the only Ideogram model offering style presets, style reference images, negative prompts and color palettes.",
      "highlights": [
        "Style references guide the look of generations",
        "Transparent-background variant available",
        "Remains available in the app and API alongside newer models"
      ],
      "source_title": "Blog Posts | Ideogram",
      "source_url": "https://docs.ideogram.ai/about-ideogram/blog-posts",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#ideogram-3-0"
    },
    {
      "slug": "xlam-2",
      "date": "2025-03-26",
      "model": "xLAM-2",
      "lab": "Salesforce",
      "lab_slug": "salesforce",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "1B, 3B, 8B, 32B and 70B",
      "model_license": "CC-BY-NC-4.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Second generation of Salesforce's large action models for function calling and AI agents, adding multi-turn conversation and tool use trained with the APIGen-MT data pipeline.",
      "highlights": [
        "Released in 1B, 3B, 8B, 32B and 70B sizes on Llama and Qwen backbones",
        "Llama-based models supported 128k context; Qwen-based models 32k, extendable to 128k",
        "Released for research use under CC BY-NC 4.0"
      ],
      "source_title": "SalesforceAIResearch/xLAM GitHub repository",
      "source_url": "https://github.com/SalesforceAIResearch/xLAM",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#xlam-2"
    },
    {
      "slug": "deepseek-v3-0324",
      "date": "2025-03-25",
      "model": "DeepSeek-V3-0324",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "DeepSeek-V3-0324 is an updated DeepSeek-V3 checkpoint with a major boost in reasoning, stronger front-end development skills, and smarter tool use. It also moved the V3 weights to the MIT License.",
      "highlights": [
        "Replaced V3 behind the deepseek-chat API model on March 24, 2025",
        "DeepSeek recommends it for non-complex reasoning tasks with DeepThink turned off",
        "Weights on Hugging Face under the MIT License"
      ],
      "source_title": "DeepSeek-V3-0324 Release",
      "source_url": "https://api-docs.deepseek.com/news/news250325",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#deepseek-v3-0324"
    },
    {
      "slug": "gemini-2-5-pro",
      "date": "2025-03-25",
      "model": "Gemini 2.5 Pro",
      "lab": "Google",
      "lab_slug": "google",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1048576,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms",
        "Consumer app"
      ],
      "summary": "Gemini 2.5 Pro was the first Gemini 2.5 model, a thinking model that reasons through its thoughts before responding, launched as an experimental release with a 1 million token context window.",
      "highlights": [
        "Available at launch in Google AI Studio and to Gemini Advanced users in the Gemini app",
        "Paid preview gemini-2.5-pro-preview-03-25 followed on Apr 4, 2025",
        "Stable gemini-2.5-pro became generally available on June 17, 2025",
        "Accepts audio, images, video, text and PDF input"
      ],
      "source_title": "Gemini 2.5: Our newest Gemini model with thinking",
      "source_url": "https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-model-thinking-updates-march-2025/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#gemini-2-5-pro"
    },
    {
      "slug": "gpt-image-1",
      "date": "2025-03-25",
      "model": "GPT-4o image generation (gpt-image-1)",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "4o image generation builds image generation natively into GPT-4o, with improved text rendering, multi-turn generation, and instruction following. It became ChatGPT's default image generator and later reached the API as gpt-image-1.",
      "highlights": [
        "Rolled out to Free, Plus, Pro, and Team users in ChatGPT and also in Sora",
        "Released in the API as gpt-image-1 on April 23, 2025",
        "Replaced DALL-E as ChatGPT's default image generator; DALL-E remained via a GPT"
      ],
      "source_title": "Introducing 4o Image Generation | OpenAI",
      "source_url": "https://openai.com/index/introducing-4o-image-generation/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#gpt-image-1"
    },
    {
      "slug": "txgemma",
      "date": "2025-03-25",
      "model": "TxGemma",
      "lab": "Google",
      "lab_slug": "google",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "2B, 9B, 27B",
      "model_license": "Health AI Developer Foundations Terms of Use",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "TxGemma is a collection of open models fine-tuned from Gemma 2 for therapeutics development, covering classification, regression and generation tasks across drug discovery. Predict variants target narrow tasks, while chat variants explain their reasoning.",
      "highlights": [
        "Fine-tuned on 7 million training examples",
        "Predict models at 2B, 9B and 27B; chat models at 9B and 27B",
        "Available on Vertex AI Model Garden and Hugging Face"
      ],
      "source_title": "Introducing TxGemma: Open models to improve therapeutics development",
      "source_url": "https://developers.googleblog.com/en/introducing-txgemma-open-models-improving-therapeutics-development/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#txgemma"
    },
    {
      "slug": "qwen2-5-vl-32b",
      "date": "2025-03-24",
      "model": "Qwen2.5-VL-32B",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32768,
      "parameters": "32B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Qwen2.5-VL-32B-Instruct is a 32B vision-language model in the Qwen2.5-VL series, further optimized with reinforcement learning. Its responses align more closely with human preferences.",
      "highlights": [
        "Stronger mathematical reasoning and fine-grained image understanding",
        "Released under the Apache 2.0 license",
        "Weights on Hugging Face and ModelScope"
      ],
      "source_title": "Qwen2.5-VL-32B: Smarter and Lighter",
      "source_url": "https://qwenlm.github.io/blog/qwen2.5-vl-32b/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#qwen2-5-vl-32b"
    },
    {
      "slug": "gpt-4o-mini-tts",
      "date": "2025-03-20",
      "model": "gpt-4o-mini-tts",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "gpt-4o-mini-tts is a text-to-speech model built on GPT-4o mini that developers can instruct on how to speak, such as tone and style, not only on what to say. It is limited to preset synthetic voices.",
      "highlights": [
        "API model ID gpt-4o-mini-tts in the text-to-speech API",
        "Steerable delivery, for example a sympathetic customer service tone",
        "Released with an Agents SDK integration for building voice agents"
      ],
      "source_title": "Introducing next-generation audio models in the API | OpenAI",
      "source_url": "https://openai.com/index/introducing-our-next-generation-audio-models/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#gpt-4o-mini-tts"
    },
    {
      "slug": "gpt-4o-transcribe",
      "date": "2025-03-20",
      "model": "gpt-4o-transcribe and gpt-4o-mini-transcribe",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio",
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 16000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "gpt-4o-transcribe and gpt-4o-mini-transcribe are speech-to-text models built on GPT-4o and GPT-4o mini. OpenAI says they transcribe more accurately than its Whisper models, especially with accents, noise, and varying speech speeds.",
      "highlights": [
        "API model IDs gpt-4o-transcribe and gpt-4o-mini-transcribe in the speech-to-text API",
        "Trained with reinforcement learning and audio-centric midtraining, per OpenAI",
        "16,000-token context window per the API model pages"
      ],
      "source_title": "Introducing next-generation audio models in the API | OpenAI",
      "source_url": "https://openai.com/index/introducing-our-next-generation-audio-models/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#gpt-4o-transcribe"
    },
    {
      "slug": "hunyuan3d-2mini-2mv",
      "date": "2025-03-18",
      "model": "Hunyuan3D-2mini and Hunyuan3D-2mv",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "image"
      ],
      "output_modalities": [
        "3d"
      ],
      "context_window": null,
      "parameters": "0.6B (2mini), 1.1B (2mv)",
      "model_license": "Tencent Hunyuan 3D 2.0 Community License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Tencent extended Hunyuan3D 2.0 with Hunyuan3D-2mini, a smaller 0.6B image-to-shape model, and Hunyuan3D-2mv, a 1.1B shape model conditioned on multiview images, followed by turbo distilled versions.",
      "highlights": [
        "Turbo step-distilled versions and FlashVDM released March 19, 2025",
        "Guidance-distilled Fast variants released with the base models"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/Hunyuan3D-2: High-Resolution 3D Assets Generation with Large Scale Hunyuan3D Diffusion Models.",
      "source_url": "https://github.com/Tencent-Hunyuan/Hunyuan3D-2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#hunyuan3d-2mini-2mv"
    },
    {
      "slug": "gr00t-n1",
      "date": "2025-03-18",
      "model": "Isaac GR00T N1",
      "lab": "NVIDIA",
      "lab_slug": "nvidia",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "action"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "NVIDIA's first open, customizable foundation model for generalized humanoid robot reasoning and skills, using a dual-system design for planning and fast motor actions.",
      "highlights": [
        "System 2 vision language model plans; System 1 generates continuous actions",
        "First of a planned family of robot foundation models",
        "Followed by GR00T N1.5, N1.6 and N1.7"
      ],
      "source_title": "NVIDIA Announces Isaac GR00T N1, the World's First Open Humanoid Robot Foundation Model, and Simulation Frameworks to Speed Robot Development | NVIDIA Newsroom",
      "source_url": "https://nvidianews.nvidia.com/news/nvidia-isaac-gr00t-n1-open-humanoid-robot-foundation-model-simulation-frameworks",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#gr00t-n1"
    },
    {
      "slug": "llama-nemotron",
      "date": "2025-03-18",
      "model": "Llama Nemotron Nano and Super",
      "lab": "NVIDIA",
      "lab_slug": "nvidia",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "8B and 49B",
      "model_license": "NVIDIA Open Model License",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Open reasoning models post-trained by NVIDIA from Meta Llama models, with Nano derived from Llama 3.1 8B and Super from Llama 3.3 70B through neural architecture search.",
      "highlights": [
        "Offered as hosted APIs on build.nvidia.com and as weights on Hugging Face",
        "A larger Ultra model was announced for multi-GPU servers",
        "Supports reasoning and chat use cases"
      ],
      "source_title": "NVIDIA Launches Family of Open Reasoning AI Models for Developers and Enterprises to Build Agentic AI Platforms | NVIDIA Newsroom",
      "source_url": "https://nvidianews.nvidia.com/news/nvidia-launches-family-of-open-reasoning-ai-models-for-developers-and-enterprises-to-build-agentic-ai-platforms",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#llama-nemotron"
    },
    {
      "slug": "stable-virtual-camera",
      "date": "2025-03-18",
      "model": "Stable Virtual Camera",
      "lab": "Stability AI",
      "lab_slug": "stability-ai",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Stability AI Non-Commercial Research Community License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Multi-view video generation model that turns 2D images into immersive videos with 3D camera control, keeping 3D consistency across sequences up to 1,000 frames.",
      "highlights": [
        "Research release under a non-commercial license",
        "Weights on Hugging Face and code on GitHub",
        "Supports seamless loops and revisiting viewpoints"
      ],
      "source_title": "Introducing Stable Virtual Camera: Multi-View Video Generation with 3D Camera Control",
      "source_url": "https://stability.ai/news-updates/introducing-stable-virtual-camera-multi-view-video-generation-with-3d-camera-control",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#stable-virtual-camera"
    },
    {
      "slug": "mistral-small-3-1",
      "date": "2025-03-17",
      "model": "Mistral Small 3.1",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "24B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Successor to Mistral Small 3 that adds image understanding and expands the context window to 128K tokens while keeping the 24B Apache 2.0 open-weight design.",
      "highlights": [
        "Base and instruct checkpoints released",
        "Improved text performance over Mistral Small 3",
        "API model ID mistral-small-2503"
      ],
      "source_title": "Mistral Small 3.1 | Mistral AI",
      "source_url": "https://mistral.ai/news/mistral-small-3-1",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#mistral-small-3-1"
    },
    {
      "slug": "step-video-ti2v",
      "date": "2025-03-17",
      "model": "Step-Video-TI2V",
      "lab": "StepFun",
      "lab_slug": "stepfun",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": "30B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Step-Video-TI2V is a 30B-parameter text-driven image-to-video model built on Step-Video-T2V, generating videos of up to 102 frames from an input image and text prompt.",
      "highlights": [
        "Generates videos up to 102 frames",
        "Released with the Step-Video-TI2V-Eval benchmark and ComfyUI integration",
        "Weights on Hugging Face under MIT"
      ],
      "source_title": "GitHub - stepfun-ai/Step-Video-TI2V",
      "source_url": "https://github.com/stepfun-ai/Step-Video-TI2V",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#step-video-ti2v"
    },
    {
      "slug": "ernie-4-5",
      "date": "2025-03-16",
      "model": "ERNIE 4.5",
      "lab": "Baidu",
      "lab_slug": "baidu",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "ERNIE 4.5 is Baidu's new-generation native multimodal foundation model, jointly modeling text, images, audio, and video, with stronger language understanding, generation, logic, and memory and reduced hallucination.",
      "highlights": [
        "Free on the ERNIE Bot website at launch",
        "API available to enterprises and developers on Baidu AI Cloud Qianfan",
        "Uses FlashMask dynamic attention masking and heterogeneous multimodal experts"
      ],
      "source_title": "两连发！文心大模型4.5及X1，免费！",
      "source_url": "https://home.baidu.com/home/index/news_detail/id/18020",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#ernie-4-5"
    },
    {
      "slug": "ernie-x1",
      "date": "2025-03-16",
      "model": "ERNIE X1",
      "lab": "Baidu",
      "lab_slug": "baidu",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "ERNIE X1 is Baidu's deep-thinking reasoning model, released with ERNIE 4.5 and described as the first deep-thinking model to use tools autonomously, including search, document Q&A, image understanding, and a code interpreter.",
      "highlights": [
        "Free to users on the ERNIE Bot website at launch",
        "Qianfan API availability announced as coming soon",
        "Trained with progressive reinforcement learning and end-to-end chain-of-thought and action training"
      ],
      "source_title": "两连发！文心大模型4.5及X1，免费！",
      "source_url": "https://home.baidu.com/home/index/news_detail/id/18020",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#ernie-x1"
    },
    {
      "slug": "command-a",
      "date": "2025-03-13",
      "model": "Command A",
      "lab": "Cohere",
      "lab_slug": "cohere",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "111B",
      "model_license": "CC-BY-NC-4.0",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "Enterprise generative model optimized for tool use, RAG, agents and multilingual work that aims for strong performance on minimal hardware.",
      "highlights": [
        "256K-token context length",
        "Supports 23 languages",
        "Research weights released as c4ai-command-a-03-2025"
      ],
      "source_title": "Introducing Command A: Max performance, minimal compute | Cohere Blog",
      "source_url": "https://cohere.com/blog/command-a",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#command-a"
    },
    {
      "slug": "olmo-2-32b",
      "date": "2025-03-13",
      "model": "OLMo 2 32B",
      "lab": "Ai2",
      "lab_slug": "ai2",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "32B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Largest model in the OLMo 2 family and a fully open model trained on up to 6 trillion tokens and post-trained with Tülu 3.1.",
      "highlights": [
        "Trained with the new OLMo-core codebase",
        "Trained on Google Cloud's Augusta AI Hypercomputer",
        "Released under Apache 2.0"
      ],
      "source_title": "OLMo 2 32B: First fully open model to outperform GPT 3.5 and GPT 4o mini | Ai2",
      "source_url": "https://allenai.org/blog/olmo2-32b",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#olmo-2-32b"
    },
    {
      "slug": "gemini-2-0-flash-image-generation",
      "date": "2025-03-12",
      "model": "Gemini 2.0 Flash native image generation",
      "lab": "Google",
      "lab_slug": "google",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text",
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Google opened Gemini 2.0 Flash's native image output to developer experimentation, letting the model generate and edit images conversationally alongside text after first offering the capability to trusted testers in December 2024.",
      "highlights": [
        "Experimental model available in Google AI Studio and the Gemini API in all supported regions",
        "Supports multi-turn conversational image editing",
        "Preview model gemini-2.0-flash-preview-image-generation followed on May 7, 2025"
      ],
      "source_title": "Experiment with Gemini 2.0 Flash native image generation",
      "source_url": "https://developers.googleblog.com/experiment-with-gemini-20-flash-native-image-generation/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#gemini-2-0-flash-image-generation"
    },
    {
      "slug": "gemma-3",
      "date": "2025-03-12",
      "model": "Gemma 3",
      "lab": "Google",
      "lab_slug": "google",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 131072,
      "parameters": "1B, 4B, 12B, 27B",
      "model_license": "Gemma Terms of Use",
      "available_at_launch": [
        "Open weights",
        "API",
        "Cloud platforms"
      ],
      "summary": "Gemma 3 is Google's third generation of open models in 1B, 4B, 12B and 27B sizes, adding image and short video input to the 4B and larger models. It expands the context window to 128K tokens.",
      "highlights": [
        "128K-token context for 4B, 12B and 27B; 32K for 1B",
        "Built-in support for over 35 languages and pretrained support for over 140",
        "Function calling and structured output",
        "Official quantized versions released"
      ],
      "source_title": "Introducing Gemma 3: The most capable model you can run on a single GPU or TPU",
      "source_url": "https://blog.google/technology/developers/gemma-3/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#gemma-3"
    },
    {
      "slug": "kwaipilot-oasis-code-embedding-1-5b",
      "date": "2025-03-12",
      "model": "OASIS-code-embedding-1.5B",
      "lab": "Kuaishou",
      "lab_slug": "kuaishou",
      "type": "embedding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "1.5B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "OASIS-code-embedding-1.5B is a code embedding model from Kuaishou's Kwaipilot team, designed to improve semantic code retrieval and search across programming contexts.",
      "highlights": [
        "Built for code retrieval and code search systems",
        "Released with a preprint on the same day",
        "Weights on Hugging Face under MIT"
      ],
      "source_title": "Kwaipilot/OASIS-code-embedding-1.5B · Hugging Face",
      "source_url": "https://huggingface.co/Kwaipilot/OASIS-code-embedding-1.5B",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#kwaipilot-oasis-code-embedding-1-5b"
    },
    {
      "slug": "shieldgemma-2",
      "date": "2025-03-12",
      "model": "ShieldGemma 2",
      "lab": "Google",
      "lab_slug": "google",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "4B",
      "model_license": "Gemma Terms of Use",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "ShieldGemma 2 is a 4B image safety classifier built on Gemma 3, released alongside it. It labels images against safety policies so developers can filter inputs and outputs of image models.",
      "highlights": [
        "Covers dangerous content, sexually explicit content and violence",
        "Trained on Gemma 3's 4B instruction-tuned checkpoint",
        "Outputs Yes/No probability scores per policy"
      ],
      "source_title": "Introducing Gemma 3: The most capable model you can run on a single GPU or TPU",
      "source_url": "https://blog.google/technology/developers/gemma-3/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#shieldgemma-2"
    },
    {
      "slug": "gpt-4o-search-preview",
      "date": "2025-03-11",
      "model": "GPT-4o Search Preview and GPT-4o mini Search Preview",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "GPT-4o Search Preview and GPT-4o mini Search Preview are fine-tuned web search models offered directly in the Chat Completions API, launched alongside the Responses API and its built-in web search tool.",
      "highlights": [
        "API model IDs gpt-4o-search-preview and gpt-4o-mini-search-preview",
        "Web search in the API uses the same model as ChatGPT search, per OpenAI",
        "Responses include inline citations linking to sources",
        "128,000-token context window per the API model pages"
      ],
      "source_title": "New tools for building agents | OpenAI",
      "source_url": "https://openai.com/index/new-tools-for-building-agents/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#gpt-4o-search-preview"
    },
    {
      "slug": "reka-flash-3",
      "date": "2025-03-10",
      "model": "Reka Flash 3",
      "lab": "Reka",
      "lab_slug": "reka",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "21B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Consumer app",
        "Open weights"
      ],
      "summary": "An open research preview of Reka's 21B general-purpose reasoning model, trained from scratch and refined with RLOO reinforcement learning, which thinks inside reasoning tags and supports budget forcing to cap thinking length.",
      "highlights": [
        "Released with a 32k context length",
        "Weights on Hugging Face under Apache-2.0, in a Llama-compatible format",
        "Primarily an English model; available to try in Reka Space",
        "An improved version powers Reka Nexus, Reka's platform for AI workers"
      ],
      "source_title": "Reasoning with Reka Flash 3",
      "source_url": "https://reka.ai/news/introducing-reka-flash",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#reka-flash-3"
    },
    {
      "slug": "gemini-embedding-001",
      "date": "2025-03-07",
      "model": "Gemini Embedding (gemini-embedding-001)",
      "lab": "Google",
      "lab_slug": "google",
      "type": "embedding",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "embedding"
      ],
      "context_window": 2048,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms"
      ],
      "summary": "Gemini Embedding is Google's first text embedding model trained on the Gemini model itself, released experimentally in March 2025 and as the stable gemini-embedding-001 in July 2025 for retrieval, classification and RAG.",
      "highlights": [
        "Supports over 100 languages",
        "Default 3072-dimension output that can be scaled down with Matryoshka Representation Learning",
        "The experimental gemini-embedding-exp-03-07 accepted 8K input tokens; the GA model supports 2,048",
        "Generally available in the Gemini API and Vertex AI on July 14, 2025"
      ],
      "source_title": "State-of-the-art text embedding via the Gemini API",
      "source_url": "https://developers.googleblog.com/gemini-embedding-text-model-now-available-gemini-api/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#gemini-embedding-001"
    },
    {
      "slug": "chirp-3",
      "date": "2025-03-06",
      "model": "Chirp 3",
      "lab": "Google",
      "lab_slug": "google",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "audio",
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms"
      ],
      "summary": "Chirp 3 is Google Cloud's latest generation of speech models, covering Chirp 3: HD voices for text-to-speech and Chirp 3: Transcription for speech recognition. HD voices expanded to 8 speakers in 31 locales in March 2025.",
      "highlights": [
        "Chirp 3: HD voices with 8 speakers and 31 locales generally available from April 2, 2025",
        "Instant Custom Voice creates a custom voice from 10 seconds of audio (allowlist)",
        "Chirp 3: Transcription generally available from October 13, 2025 with diarization and automatic language detection"
      ],
      "source_title": "Cloud TTS release notes",
      "source_url": "https://docs.cloud.google.com/text-to-speech/docs/release-notes",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#chirp-3"
    },
    {
      "slug": "hunyuanvideo-i2v",
      "date": "2025-03-06",
      "model": "HunyuanVideo-I2V",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Tencent Hunyuan Community License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "HunyuanVideo-I2V is an open image-to-video generation model based on HunyuanVideo, released with inference code, weights, and LoRA training code for customizable video effects.",
      "highlights": [
        "Parallel inference via xDiT added March 13, 2025",
        "LoRA effects examples include hair growth and embrace"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/HunyuanVideo-I2V: HunyuanVideo-I2V: A Customizable Image-to-Video Model based on HunyuanVideo",
      "source_url": "https://github.com/Tencent-Hunyuan/HunyuanVideo-I2V",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#hunyuanvideo-i2v"
    },
    {
      "slug": "jamba-1-6",
      "date": "2025-03-06",
      "model": "Jamba 1.6",
      "lab": "AI21 Labs",
      "lab_slug": "ai21",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": null,
      "model_license": "Jamba Open Model License",
      "available_at_launch": [
        "Open weights",
        "API",
        "Enterprise"
      ],
      "summary": "Jamba Large 1.6 and Jamba Mini 1.6, an update of AI21's open hybrid SSM-Transformer family aimed at long-context enterprise deployment.",
      "highlights": [
        "256K-token context window",
        "Deployable privately in VPC or on-premises",
        "Weights on Hugging Face"
      ],
      "source_title": "AI21's Jamba 1.6: The Best Open Model for Private Enterprise Deployment",
      "source_url": "https://www.ai21.com/blog/introducing-jamba-1-6/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#jamba-1-6"
    },
    {
      "slug": "mistral-ocr",
      "date": "2025-03-06",
      "model": "Mistral OCR",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Optical character recognition API that reads images and PDFs and returns ordered, interleaved text and images, including tables and equations, for use in RAG pipelines.",
      "highlights": [
        "Became the default document understanding model in Le Chat",
        "API model ID mistral-ocr-2503",
        "Since deprecated in favor of newer OCR models"
      ],
      "source_title": "Mistral OCR | Mistral AI",
      "source_url": "https://mistral.ai/news/mistral-ocr",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#mistral-ocr"
    },
    {
      "slug": "qwq-32b",
      "date": "2025-03-06",
      "model": "QwQ-32B",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 131072,
      "parameters": "32B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Consumer app",
        "Open weights"
      ],
      "summary": "QwQ-32B is Qwen's 32B reasoning model trained with scaled reinforcement learning, which Qwen says performs comparably to the much larger DeepSeek-R1. It integrates agent capabilities for reasoning with tools.",
      "highlights": [
        "Full 131,072-token context length",
        "Accessible via Qwen Chat",
        "Weights on Hugging Face and ModelScope under the Apache 2.0 license"
      ],
      "source_title": "QwQ-32B: Embracing the Power of Reinforcement Learning",
      "source_url": "https://qwenlm.github.io/blog/qwq-32b/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#qwq-32b"
    },
    {
      "slug": "ltx-video-0-9-5",
      "date": "2025-03-05",
      "model": "LTX-Video 0.9.5",
      "lab": "Lightricks",
      "lab_slug": "lightricks",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": "2B",
      "model_license": "OpenRAIL-M",
      "available_at_launch": [
        "Open weights",
        "Consumer app"
      ],
      "summary": "LTX-Video checkpoint with improved quality, keyframe and video extension support, higher resolutions, better prompt understanding and an improved VAE, moving to a license that permits commercial use.",
      "highlights": [
        "New OpenRAIL-M license allowing commercial use",
        "Keyframes and video extension",
        "Online web app in LTX Studio"
      ],
      "source_title": "Lightricks/LTX-Video",
      "source_url": "https://github.com/Lightricks/LTX-Video",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#ltx-video-0-9-5"
    },
    {
      "slug": "aya-vision",
      "date": "2025-03-04",
      "model": "Aya Vision",
      "lab": "Cohere",
      "lab_slug": "cohere",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "8B and 32B",
      "model_license": "CC-BY-NC-4.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Multilingual vision-language models from Cohere For AI that extend multimodal capabilities to 23 languages for captioning, visual question answering and translation.",
      "highlights": [
        "Released in 8B and 32B sizes",
        "Uses synthetic annotations and multimodal model merging",
        "Weights published under a non-commercial license"
      ],
      "source_title": "Cohere For AI Launches Aya Vision, a Multilingual Multimodal AI Model",
      "source_url": "https://cohere.com/blog/aya-vision",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#aya-vision"
    },
    {
      "slug": "cogview4",
      "date": "2025-03-04",
      "model": "CogView4",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "6B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "CogView4 is Z.ai's open-weights text-to-image model with 6B parameters that supports native Chinese input and Chinese text-to-image generation. Z.ai released it in a diffusers-compatible version with an online demo.",
      "highlights": [
        "Native Chinese prompt input",
        "Diffusers version open-sourced on March 4, 2025",
        "Weights under the Apache 2.0 license"
      ],
      "source_title": "CogView4 & CogView3 & CogView-3Plus (GitHub)",
      "source_url": "https://github.com/zai-org/CogView4",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/03#cogview4"
    },
    {
      "slug": "command-r7b-arabic",
      "date": "2025-02-27",
      "model": "Command R7B Arabic",
      "lab": "Cohere",
      "lab_slug": "cohere",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "7B",
      "model_license": "CC-BY-NC-4.0",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "Version of Command R7B optimized for Arabic language and culture for enterprises in the Middle East and North Africa, released with open weights.",
      "highlights": [
        "Runs on low-end GPUs, a MacBook or CPUs",
        "128K-token context length",
        "Improves on all Arabic language dimensions versus Command R7B"
      ],
      "source_title": "Introducing Command R7B Arabic",
      "source_url": "https://cohere.com/blog/command-r7b-arabic",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/02#command-r7b-arabic"
    },
    {
      "slug": "gpt-4-5",
      "date": "2025-02-27",
      "model": "GPT-4.5 (research preview)",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "GPT-4.5 is a research preview of OpenAI's largest chat model at the time, built by scaling unsupervised pre-training and post-training. OpenAI cites broader knowledge, better grasp of user intent, and fewer hallucinations, without step-by-step reasoning.",
      "highlights": [
        "API model ID gpt-4.5-preview in the Chat Completions, Assistants, and Batch APIs",
        "Available first to ChatGPT Pro users, with Plus and Team the following week",
        "Supports function calling, Structured Outputs, streaming, and image input",
        "128,000-token context window per the API model page"
      ],
      "source_title": "Introducing GPT-4.5 | OpenAI",
      "source_url": "https://openai.com/index/introducing-gpt-4-5/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/02#gpt-4-5"
    },
    {
      "slug": "ideogram-2a",
      "date": "2025-02-27",
      "model": "Ideogram 2a",
      "lab": "Ideogram",
      "lab_slug": "ideogram",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Ideogram's fastest and most affordable text-to-image model at launch, optimized for graphic design and photography and released on the website, API and partner platforms.",
      "highlights": [
        "Generates images in 10 seconds, or 5 seconds with 2a Turbo",
        "Launched with image prices cut by 50% on the website and API",
        "Offered through partners including Replicate, fal and Freepik"
      ],
      "source_title": "Ideogram on X: Say hello to Ideogram 2a, our fastest and most affordable text-to-image model to date",
      "source_url": "https://x.com/ideogram_ai/status/1895157668102222075",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/02#ideogram-2a"
    },
    {
      "slug": "granite-3-2",
      "date": "2025-02-26",
      "model": "Granite 3.2",
      "lab": "IBM",
      "lab_slug": "ibm",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "2B and 8B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Update of the Granite language models that adds optional thinking capabilities to the 2B and 8B instruct models, built on Granite 3.1 Instruct.",
      "highlights": [
        "Reasoning can be toggled on or off",
        "Built on Granite 3.1 8B Instruct",
        "Released alongside Granite Vision 3.2 and Granite Guardian 3.2"
      ],
      "source_title": "ibm-granite/granite-3.2-8b-instruct · Hugging Face",
      "source_url": "https://huggingface.co/ibm-granite/granite-3.2-8b-instruct",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/02#granite-3-2"
    },
    {
      "slug": "granite-guardian-3-2",
      "date": "2025-02-26",
      "model": "Granite Guardian 3.2",
      "lab": "IBM",
      "lab_slug": "ibm",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "5B and 3B-A800M",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Smaller Granite Guardian safety models, including a 5B model pruned and healed from Granite Guardian 3.1 8B and a 3B mixture-of-experts model with 800M active parameters.",
      "highlights": [
        "About 30% of the 8B model's parameters removed for the 5B version",
        "Detects risks in prompts and responses",
        "Released under Apache 2.0"
      ],
      "source_title": "ibm-granite/granite-guardian-3.2-3b-a800m · Hugging Face",
      "source_url": "https://huggingface.co/ibm-granite/granite-guardian-3.2-3b-a800m",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/02#granite-guardian-3-2"
    },
    {
      "slug": "granite-vision-3-2",
      "date": "2025-02-26",
      "model": "Granite Vision 3.2",
      "lab": "IBM",
      "lab_slug": "ibm",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "2B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Compact IBM vision-language model designed for visual document understanding, automating content extraction from tables, charts, infographics, plots and diagrams for enterprise use.",
      "highlights": [
        "Follows the Granite Vision 3.1 preview from January 2025",
        "Focused on enterprise document tasks",
        "Released under Apache 2.0"
      ],
      "source_title": "ibm-granite/granite-vision-3.2-2b · Hugging Face",
      "source_url": "https://huggingface.co/ibm-granite/granite-vision-3.2-2b",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/02#granite-vision-3-2"
    },
    {
      "slug": "phi-4-mini",
      "date": "2025-02-26",
      "model": "Phi-4-mini",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "3.8B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights",
        "API",
        "Cloud platforms"
      ],
      "summary": "Phi-4-mini is a 3.8B dense, decoder-only Microsoft language model with grouped-query attention and a 200,000-token vocabulary, designed for speed and efficiency, with support for function calling and long inputs.",
      "highlights": [
        "Supports sequences up to 128,000 tokens",
        "200,000-token vocabulary with grouped-query attention",
        "Available in Azure AI Foundry, Hugging Face and the NVIDIA API Catalog"
      ],
      "source_title": "Empowering innovation: The next generation of the Phi family",
      "source_url": "https://azure.microsoft.com/en-us/blog/empowering-innovation-the-next-generation-of-the-phi-family/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/02#phi-4-mini"
    },
    {
      "slug": "phi-4-multimodal",
      "date": "2025-02-26",
      "model": "Phi-4-multimodal",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "5.6B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights",
        "API",
        "Cloud platforms"
      ],
      "summary": "Phi-4-multimodal is Microsoft's first multimodal Phi language model, a 5.6B model that processes speech, vision and text in a single unified model rather than a pipeline of separate models.",
      "highlights": [
        "5.6B parameters with a 128K token context length",
        "Handles speech, vision and text in one model",
        "Available in Azure AI Foundry, Hugging Face and the NVIDIA API Catalog"
      ],
      "source_title": "Empowering innovation: The next generation of the Phi family",
      "source_url": "https://azure.microsoft.com/en-us/blog/empowering-innovation-the-next-generation-of-the-phi-family/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/02#phi-4-multimodal"
    },
    {
      "slug": "elevenlabs-scribe-v1",
      "date": "2025-02-26",
      "model": "Scribe v1",
      "lab": "ElevenLabs",
      "lab_slug": "elevenlabs",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "ElevenLabs' first speech-to-text model, transcribing speech in 99 languages with word-level timestamps, speaker diarization and audio-event tagging in a structured response.",
      "highlights": [
        "99 languages",
        "Word-level timestamps and speaker diarization",
        "Audio-event tagging in a structured response"
      ],
      "source_title": "Meet Scribe",
      "source_url": "https://elevenlabs.io/blog/meet-scribe",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/02#elevenlabs-scribe-v1"
    },
    {
      "slug": "olmocr",
      "date": "2025-02-25",
      "model": "olmOCR",
      "lab": "Ai2",
      "lab_slug": "ai2",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "7B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Open PDF text extraction toolkit built on a fine-tuned Qwen2-VL-7B-Instruct model, released with weights, fine-tuning data and an efficient inference pipeline.",
      "highlights": [
        "Built on Qwen2-VL-7B-Instruct",
        "Released with training and inference code",
        "Followed by olmOCR 2 in October 2025"
      ],
      "source_title": "olmOCR: Efficient PDF text extraction with vision language models | Ai2",
      "source_url": "https://allenai.org/blog/olmocr",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/02#olmocr"
    },
    {
      "slug": "qwq-max-preview",
      "date": "2025-02-25",
      "model": "QwQ-Max-Preview",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "QwQ-Max-Preview is a preview of Qwen's QwQ-Max reasoning model, built on Qwen2.5-Max and strong in math, coding, and agent tasks. Qwen said it planned to open-source QwQ-Max and Qwen2.5-Max under Apache 2.0.",
      "highlights": [
        "Available on Qwen Chat",
        "Announced alongside plans for a Qwen Chat app",
        "Qwen also planned smaller open reasoning models such as QwQ-32B"
      ],
      "source_title": "QwQ-Max-Preview",
      "source_url": "https://qwenlm.github.io/blog/qwq-max-preview/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/02#qwq-max-preview"
    },
    {
      "slug": "wan2-1",
      "date": "2025-02-25",
      "model": "Wan2.1",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": "1.3B, 14B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Wan2.1 is Alibaba's open video generation model family, released with inference code and weights for text-to-video and image-to-video at 1.3B and 14B sizes.",
      "highlights": [
        "Text-to-video in 1.3B and 14B; image-to-video 14B at 480P and 720P",
        "First-last-frame-to-video model Wan2.1-FLF2V-14B followed April 17, 2025",
        "Released under the Apache 2.0 license"
      ],
      "source_title": "Wan2.1 (GitHub)",
      "source_url": "https://github.com/Wan-Video/Wan2.1",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/02#wan2-1"
    },
    {
      "slug": "claude-3-7-sonnet",
      "date": "2025-02-24",
      "model": "Claude 3.7 Sonnet",
      "lab": "Anthropic",
      "lab_slug": "anthropic",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Cloud platforms",
        "Enterprise"
      ],
      "summary": "Claude 3.7 Sonnet is Anthropic's first hybrid reasoning model, able to give near-instant responses or visible step-by-step extended thinking. Anthropic introduced Claude Code, a command line tool for agentic coding, as a research preview alongside it.",
      "highlights": [
        "API model ID claude-3-7-sonnet-20250219",
        "API users can set a thinking budget up to the 128K-token output limit",
        "Available on all Claude plans; extended thinking excluded from the free tier",
        "Available on Amazon Bedrock and Google Cloud Vertex AI"
      ],
      "source_title": "Claude 3.7 Sonnet and Claude Code",
      "source_url": "https://www.anthropic.com/news/claude-3-7-sonnet",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/02#claude-3-7-sonnet"
    },
    {
      "slug": "moonlight",
      "date": "2025-02-23",
      "model": "Moonlight",
      "lab": "Moonshot AI",
      "lab_slug": "moonshot-ai",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "MoE: 16B total, 3B active",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Moonlight is a 3B-active, 16B-total mixture-of-experts model trained on 5.7T tokens with the Muon optimizer. Moonshot released it to show that Muon scales to large-model training without hyperparameter tuning.",
      "highlights": [
        "Moonshot reports Muon is about 2 times more sample efficient than Adam",
        "Released pretrained, instruction-tuned, and intermediate checkpoints",
        "Muon implementation open-sourced alongside the model"
      ],
      "source_title": "Moonlight: Muon is Scalable for LLM Training (GitHub)",
      "source_url": "https://github.com/MoonshotAI/Moonlight",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/02#moonlight"
    },
    {
      "slug": "smolvlm2",
      "date": "2025-02-20",
      "model": "SmolVLM2",
      "lab": "Hugging Face",
      "lab_slug": "hugging-face",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "256M, 500M and 2.2B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Video-capable SmolVLM models in three sizes that bring visual and video understanding to small devices, with the 500M and 256M versions billed as the smallest video language models.",
      "highlights": [
        "MLX-ready with Python and Swift APIs from launch",
        "2.2B model aimed at vision and video tasks",
        "Released under Apache 2.0"
      ],
      "source_title": "SmolVLM2: Bringing Video Understanding to Every Device",
      "source_url": "https://huggingface.co/blog/smolvlm2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/02#smolvlm2"
    },
    {
      "slug": "grok-3",
      "date": "2025-02-19",
      "model": "Grok 3 and Grok 3 mini",
      "lab": "xAI",
      "lab_slug": "xai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Grok 3 is xAI's flagship model trained on the Colossus supercluster, launched in beta alongside Grok 3 mini, with Think reasoning modes refined through large-scale reinforcement learning.",
      "highlights": [
        "Rolled out to X Premium and Premium+ users on X and Grok.com",
        "Announcement states a 1M-token context window for Grok 3",
        "Launched alongside DeepSearch, xAI's first agent",
        "Grok 3 models became generally available on the API on April 3, 2025"
      ],
      "source_title": "Grok 3 Beta: The Age of Reasoning Agents | SpaceXSI",
      "source_url": "https://x.ai/news/grok-3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/02#grok-3"
    },
    {
      "slug": "paligemma-2-mix",
      "date": "2025-02-19",
      "model": "PaliGemma 2 mix",
      "lab": "Google",
      "lab_slug": "google",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "3B, 10B, 28B",
      "model_license": "Gemma Terms of Use",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "PaliGemma 2 mix is a set of PaliGemma 2 checkpoints fine-tuned on a mixture of tasks, so one model handles captioning, OCR, visual question answering, object detection and segmentation, selected through the prompt.",
      "highlights": [
        "Available at 224px and 448px resolutions",
        "Drop-in upgrade for original PaliGemma mix users",
        "Weights on Kaggle and Hugging Face; deployable in Vertex Model Garden"
      ],
      "source_title": "Introducing PaliGemma 2 mix: A vision-language model for multiple tasks",
      "source_url": "https://developers.googleblog.com/en/introducing-paligemma-2-mix/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/02#paligemma-2-mix"
    },
    {
      "slug": "r1-1776",
      "date": "2025-02-18",
      "model": "R1 1776",
      "lab": "Perplexity",
      "lab_slug": "perplexity",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "A version of DeepSeek-R1 that Perplexity post-trained on about 40,000 multilingual prompts covering topics censored by the Chinese Communist Party, aiming for factual answers while keeping R1's reasoning ability.",
      "highlights": [
        "Weights released on Hugging Face under the MIT license",
        "Also offered through Perplexity's Sonar API at launch",
        "Post-trained with an adapted version of NVIDIA's NeMo 2.0 framework",
        "Removed from the Perplexity API on August 1, 2025, per the API changelog"
      ],
      "source_title": "Open-sourcing R1 1776",
      "source_url": "https://web.archive.org/web/20250219025612/https://www.perplexity.ai/hub/blog/open-sourcing-r1-1776",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/02#r1-1776"
    },
    {
      "slug": "mistral-saba",
      "date": "2025-02-17",
      "model": "Mistral Saba",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32000,
      "parameters": "24B",
      "model_license": null,
      "available_at_launch": [
        "API",
        "Enterprise"
      ],
      "summary": "First of Mistral's specialized regional language models, a 24B model trained on curated data from the Middle East and South Asia for Arabic and Indian-origin languages.",
      "highlights": [
        "Particularly strong in South Indian-origin languages such as Tamil",
        "Available through the API and for local deployment",
        "API model ID mistral-saba-2502"
      ],
      "source_title": "Mistral Saba | Mistral AI",
      "source_url": "https://mistral.ai/news/mistral-saba",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/02#mistral-saba"
    },
    {
      "slug": "step-audio",
      "date": "2025-02-17",
      "model": "Step-Audio",
      "lab": "StepFun",
      "lab_slug": "stepfun",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "audio",
        "text"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": null,
      "parameters": "130B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Step-Audio is StepFun's open speech interaction family, led by the 130B-parameter Step-Audio-Chat model that unifies speech understanding and generation, released with a 3B text-to-speech model and an audio tokenizer.",
      "highlights": [
        "Step-Audio-Chat is a 130B-parameter unified speech model",
        "Step-Audio-TTS-3B and Step-Audio-Tokenizer released alongside",
        "Released with the StepEval-Audio-360 multi-turn benchmark"
      ],
      "source_title": "GitHub - stepfun-ai/Step-Audio",
      "source_url": "https://github.com/stepfun-ai/Step-Audio",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/02#step-audio"
    },
    {
      "slug": "step-video-t2v",
      "date": "2025-02-17",
      "model": "Step-Video-T2V",
      "lab": "StepFun",
      "lab_slug": "stepfun",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": "30B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Step-Video-T2V is StepFun's 30B-parameter open text-to-video model, capable of generating videos of up to 204 frames, released alongside a faster Turbo variant and a technical report.",
      "highlights": [
        "Generates videos up to 204 frames",
        "Step-Video-T2V-Turbo released the same day",
        "Weights on Hugging Face under MIT"
      ],
      "source_title": "GitHub - stepfun-ai/Step-Video-T2V",
      "source_url": "https://github.com/stepfun-ai/Step-Video-T2V",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/02#step-video-t2v"
    },
    {
      "slug": "minimax-image-01",
      "date": "2025-02-15",
      "model": "MiniMax Image-01",
      "lab": "MiniMax",
      "lab_slug": "minimax",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Image-01 is MiniMax's first text-to-image generation model, offered through the MiniMax API platform with an emphasis on prompt adherence, lighting, detailed environments, and realistic people and objects.",
      "highlights": [
        "API release notes list it on February 15, 2025; the announcement post is dated February 28, 2025",
        "Supports multiple standard aspect ratios",
        "Later joined by image-01-live for generation from a reference image"
      ],
      "source_title": "MiniMax Launches Image-01: Expanding Multimodal Vision with Text-to-Image Generation",
      "source_url": "https://www.minimax.io/news/image-01",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/02#minimax-image-01"
    },
    {
      "slug": "deephermes-3-preview",
      "date": "2025-02-14",
      "model": "DeepHermes-3-Llama-3-8B-Preview",
      "lab": "Nous Research",
      "lab_slug": "nous-research",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "8B",
      "model_license": "Llama 3 Community License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Preview Hermes model from Nous Research that unifies deep reasoning and standard chat modes in a single model, built on Meta's Llama 3 8B.",
      "highlights": [
        "Switches between deep reasoning and chat modes",
        "3B and Mistral 24B preview versions also released",
        "Built on Llama 3 8B"
      ],
      "source_title": "Releases | Nous Research",
      "source_url": "https://nousresearch.com/releases",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/02#deephermes-3-preview"
    },
    {
      "slug": "minimax-video-01-director",
      "date": "2025-02-11",
      "model": "MiniMax T2V-01-Director and I2V-01-Director",
      "lab": "MiniMax",
      "lab_slug": "minimax",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "T2V-01-Director and I2V-01-Director are Hailuo text-to-video and image-to-video models that follow camera movement instructions more precisely, reduce movement randomness, and offer preset camera settings for cinematic shots.",
      "highlights": [
        "API release notes list the release on February 11, 2025; the announcement post is dated March 3, 2025",
        "Available on Hailuo AI web and mobile and the MiniMax API"
      ],
      "source_title": "Hailuo AI Advances Cinematic Storytelling with T2V-01-Director and I2V-01-Director",
      "source_url": "https://www.minimax.io/news/01-director",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/02#minimax-video-01-director"
    },
    {
      "slug": "perplexity-sonar-2025-02",
      "date": "2025-02-11",
      "model": "Sonar (built on Llama 3.3 70B)",
      "lab": "Perplexity",
      "lab_slug": "perplexity",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "70B (Llama 3.3 70B base)",
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "An updated version of Perplexity's in-house Sonar model, built on Llama 3.3 70B and further trained for answer factuality and readability in Perplexity's default search mode, served on Cerebras inference infrastructure.",
      "highlights": [
        "Available to Perplexity Pro users, who can set it as their default model",
        "Perplexity reports about 1,200 tokens per second on Cerebras infrastructure",
        "Also offered via the Sonar API, which did not yet run on Cerebras at launch"
      ],
      "source_title": "Meet New Sonar: A Blazing Fast Model Optimized for Perplexity Search",
      "source_url": "https://www.perplexity.ai/hub/blog/meet-new-sonar",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/02#perplexity-sonar-2025-02"
    },
    {
      "slug": "gemini-2-0-flash-lite",
      "date": "2025-02-05",
      "model": "Gemini 2.0 Flash-Lite",
      "lab": "Google",
      "lab_slug": "google",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms"
      ],
      "summary": "Gemini 2.0 Flash-Lite was introduced as Google's most cost-efficient model at the time, designed for speed and scale, with a 1 million token context window and multimodal input.",
      "highlights": [
        "Public preview in Google AI Studio and Vertex AI at launch",
        "Generally available in the Gemini API and Vertex AI on Feb 25, 2025",
        "Shut down in the Gemini API on June 1, 2026"
      ],
      "source_title": "Gemini 2.0 model updates: 2.0 Flash, Flash-Lite, Pro Experimental",
      "source_url": "https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-model-updates-february-2025/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/02#gemini-2-0-flash-lite"
    },
    {
      "slug": "gemini-2-0-pro",
      "date": "2025-02-05",
      "model": "Gemini 2.0 Pro",
      "lab": "Google",
      "lab_slug": "google",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 2000000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms",
        "Consumer app"
      ],
      "summary": "Gemini 2.0 Pro was released as an experimental model that Google described as its best yet for coding performance and complex prompts, with its largest context window at the time of 2 million tokens.",
      "highlights": [
        "Experimental model ID gemini-2.0-pro-exp-02-05",
        "Available in Google AI Studio, Vertex AI and to Gemini Advanced users in the Gemini app",
        "Able to call tools such as Google Search and code execution"
      ],
      "source_title": "Gemini 2.0 model updates: 2.0 Flash, Flash-Lite, Pro Experimental",
      "source_url": "https://blog.google/innovation-and-ai/models-and-research/google-deepmind/gemini-model-updates-february-2025/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/02#gemini-2-0-pro"
    },
    {
      "slug": "o3-mini",
      "date": "2025-01-31",
      "model": "OpenAI o3-mini",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 200000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "o3-mini is a small, cost-efficient reasoning model optimized for STEM tasks such as science, math, and coding. It replaced o1-mini in ChatGPT and offers low, medium, and high reasoning effort in the API.",
      "highlights": [
        "API model ID o3-mini (snapshot o3-mini-2025-01-31)",
        "First OpenAI small reasoning model with function calling, Structured Outputs, and developer messages",
        "Free ChatGPT users could try it via 'Reason'; paid users also got o3-mini-high",
        "No vision support; 200,000-token context window per the API model page"
      ],
      "source_title": "OpenAI o3-mini | OpenAI",
      "source_url": "https://openai.com/index/openai-o3-mini/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/01#o3-mini"
    },
    {
      "slug": "mistral-small-3",
      "date": "2025-01-30",
      "model": "Mistral Small 3",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32000,
      "parameters": "24B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Latency-optimized 24B model released under Apache 2.0 with pretrained and instruct checkpoints, marking Mistral's return to Apache licensing for general-purpose models.",
      "highlights": [
        "Uses far fewer layers than comparable models to cut latency",
        "Not trained with RL or synthetic data",
        "API model ID mistral-small-2501"
      ],
      "source_title": "Mistral Small 3 | Mistral AI",
      "source_url": "https://mistral.ai/news/mistral-small-3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/01#mistral-small-3"
    },
    {
      "slug": "sfr-embedding-code",
      "date": "2025-01-30",
      "model": "SFR-Embedding-Code",
      "lab": "Salesforce",
      "lab_slug": "salesforce",
      "type": "embedding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "embedding"
      ],
      "context_window": null,
      "parameters": "400M and 2B",
      "model_license": "CC-BY-NC-4.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Family of embedding models for code and text retrieval in 400M, 2B and 7B sizes, with the 400M and 2B models released openly.",
      "highlights": [
        "The 7B model ranked first on the CoIR code retrieval benchmark at announcement",
        "The 400M model was designed for faster, lower-cost inference, including on CPUs",
        "Open models released for research use under CC BY-NC 4.0"
      ],
      "source_title": "SFR-Embedding-Code: A Family of Embedding Models for Code Retrieval",
      "source_url": "https://www.salesforce.com/blog/sfr-embedding-code/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/01#sfr-embedding-code"
    },
    {
      "slug": "tulu-3-405b",
      "date": "2025-01-30",
      "model": "Tülu 3 405B",
      "lab": "Ai2",
      "lab_slug": "ai2",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "405B",
      "model_license": "Llama 3.1 Community License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Application of the fully open Tülu 3 post-training recipe to the Llama 3.1 405B base model, used to stress-test reinforcement learning with verifiable rewards at scale.",
      "highlights": [
        "Trained with reinforcement learning from verifiable rewards",
        "Served with vLLM using 16-way tensor parallelism during RL",
        "Built on Llama 3.1 405B"
      ],
      "source_title": "Scaling the Tülu 3 post-training recipes to surpass the performance of DeepSeek V3 | Ai2",
      "source_url": "https://allenai.org/blog/tulu-3-405b",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/01#tulu-3-405b"
    },
    {
      "slug": "qwen2-5-max",
      "date": "2025-01-28",
      "model": "Qwen2.5-Max",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Qwen2.5-Max is a large-scale mixture-of-experts model pre-trained on more than 20T tokens and post-trained with supervised fine-tuning and RLHF. Qwen released its API through Alibaba Cloud.",
      "highlights": [
        "Mixture-of-experts architecture",
        "API available through Alibaba Cloud Model Studio",
        "Available to try on Qwen Chat"
      ],
      "source_title": "Qwen2.5-Max: Exploring the Intelligence of Large-scale MoE Model",
      "source_url": "https://qwenlm.github.io/blog/qwen2.5-max/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/01#qwen2-5-max"
    },
    {
      "slug": "janus-pro",
      "date": "2025-01-27",
      "model": "Janus-Pro",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text",
        "image"
      ],
      "context_window": null,
      "parameters": "1B, 7B",
      "model_license": "DeepSeek Model License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Janus-Pro is an advanced version of Janus, DeepSeek's autoregressive framework that unifies multimodal understanding and image generation by decoupling visual encoding into separate pathways. It significantly improves both understanding and visual generation.",
      "highlights": [
        "Built on DeepSeek-LLM-1.5b-base and DeepSeek-LLM-7b-base",
        "Single unified transformer for image understanding and text-to-image generation",
        "Code under the MIT License; model use under the DeepSeek Model License"
      ],
      "source_title": "Janus-Series: Unified Multimodal Understanding and Generation Models (GitHub)",
      "source_url": "https://github.com/deepseek-ai/Janus",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/01#janus-pro"
    },
    {
      "slug": "qwen2-5-1m",
      "date": "2025-01-27",
      "model": "Qwen2.5-1M",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1010000,
      "parameters": "7B, 14B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Qwen2.5-1M is the first set of open Qwen models upgraded to handle 1M-token contexts, released as Qwen2.5-7B-Instruct-1M and Qwen2.5-14B-Instruct-1M with a vLLM-based inference framework.",
      "highlights": [
        "Context length of 1,010,000 tokens with 8,192-token generation",
        "Inference framework with sparse attention processes 1M-token inputs 3 to 7 times faster",
        "Weights under the Apache 2.0 license"
      ],
      "source_title": "Qwen2.5-1M: Deploy Your Own Qwen with Context Length up to 1M Tokens",
      "source_url": "https://qwenlm.github.io/blog/qwen2.5-1m/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/01#qwen2-5-1m"
    },
    {
      "slug": "qwen2-5-vl",
      "date": "2025-01-26",
      "model": "Qwen2.5-VL",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32768,
      "parameters": "3B, 7B, 72B",
      "model_license": "Apache-2.0 (7B); Qwen License Agreement (72B); Qwen Research License Agreement (3B)",
      "available_at_launch": [
        "Consumer app",
        "Open weights"
      ],
      "summary": "Qwen2.5-VL is Qwen's vision-language model series released in 3B, 7B, and 72B sizes as base and instruct models. The flagship Qwen2.5-VL-72B-Instruct covers document understanding, video understanding, and visual agent tasks.",
      "highlights": [
        "Qwen2.5-VL-72B-Instruct available on Qwen Chat",
        "Qwen2.5-VL-3B targets edge AI",
        "Weights on Hugging Face and ModelScope"
      ],
      "source_title": "Qwen2.5 VL! Qwen2.5 VL! Qwen2.5 VL!",
      "source_url": "https://qwenlm.github.io/blog/qwen2.5-vl/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/01#qwen2-5-vl"
    },
    {
      "slug": "hunyuan-7b-0124",
      "date": "2025-01-24",
      "model": "Hunyuan-7B-0124",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "7B",
      "model_license": "Tencent Hunyuan Community License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Hunyuan-7B-0124 is an open 7B dense language model released in pretrained and instruct versions, using improved data allocation and training, grouped query attention, and a context window extended to 256K.",
      "highlights": [
        "Released as Hunyuan-7B-Pretrain-0124 and Hunyuan-7B-Instruct-0124",
        "Hugging Face format with a vLLM inference backend"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/Tencent-Hunyuan-7B-0124",
      "source_url": "https://github.com/Tencent-Hunyuan/Tencent-Hunyuan-7B-0124",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/01#hunyuan-7b-0124"
    },
    {
      "slug": "computer-using-agent",
      "date": "2025-01-23",
      "model": "Computer-Using Agent (computer-use-preview)",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 8192,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Computer-Using Agent (CUA) combines GPT-4o's vision with reasoning trained through reinforcement learning to operate graphical interfaces from screenshots using a virtual mouse and keyboard. It launched powering the Operator research preview.",
      "highlights": [
        "Launched in Operator for ChatGPT Pro users in the U.S.",
        "Released in the API as computer-use-preview with the Responses API on March 11, 2025",
        "Asks for user confirmation on sensitive actions such as logins and CAPTCHAs",
        "API model page lists an 8,192-token context window"
      ],
      "source_title": "Computer-Using Agent | OpenAI",
      "source_url": "https://openai.com/index/computer-using-agent/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/01#computer-using-agent"
    },
    {
      "slug": "smolvlm-256m-500m",
      "date": "2025-01-23",
      "model": "SmolVLM 256M and 500M",
      "lab": "Hugging Face",
      "lab_slug": "hugging-face",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "256M and 500M",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Smaller additions to the SmolVLM family, with the 256M model presented as the smallest vision-language model in the world at release.",
      "highlights": [
        "Released with ColSmolVLM retrieval models",
        "Base and instruct versions",
        "Released under Apache 2.0"
      ],
      "source_title": "SmolVLM Grows Smaller – Introducing the 256M & 500M Models!",
      "source_url": "https://huggingface.co/blog/smolervlm",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/01#smolvlm-256m-500m"
    },
    {
      "slug": "doubao-1-5-pro",
      "date": "2025-01-22",
      "model": "Doubao-1.5-pro",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Doubao-1.5-pro is ByteDance's sparse mixture-of-experts flagship model, built with an integrated training and inference design that uses few active parameters, with upgraded vision and end-to-end speech capabilities.",
      "highlights": [
        "Offered on Volcano Engine in 32K and 256K context versions",
        "Released alongside Doubao-1.5-lite and Doubao-1.5-vision-pro",
        "Uses a 2.4B-parameter Doubao ViT with native dynamic resolution",
        "Post-training used no data from other models"
      ],
      "source_title": "Doubao 1.5pro - Doubao Team",
      "source_url": "https://seed.bytedance.com/en/special/doubao_1_5_pro",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/01#doubao-1-5-pro"
    },
    {
      "slug": "hunyuan3d-2-0",
      "date": "2025-01-21",
      "model": "Hunyuan3D 2.0",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "image"
      ],
      "output_modalities": [
        "3d"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Tencent Hunyuan 3D 2.0 Community License Agreement",
      "available_at_launch": [
        "Open weights",
        "Consumer app"
      ],
      "summary": "Hunyuan3D 2.0 is an open large-scale 3D asset generation system that pairs a shape generation model with a texture synthesis model to produce high-resolution textured 3D assets from images.",
      "highlights": [
        "Hunyuan3D-DiT-v2-0 shape model has 1.1B parameters and Hunyuan3D-Paint-v2-0 has 1.3B",
        "Launched alongside the Hunyuan3D Studio website"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/Hunyuan3D-2: High-Resolution 3D Assets Generation with Large Scale Hunyuan3D Diffusion Models.",
      "source_url": "https://github.com/Tencent-Hunyuan/Hunyuan3D-2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/01#hunyuan3d-2-0"
    },
    {
      "slug": "perplexity-sonar-pro",
      "date": "2025-01-21",
      "model": "Sonar and Sonar Pro (API)",
      "lab": "Perplexity",
      "lab_slug": "perplexity",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Perplexity's Sonar API models with built-in real-time web search and citations: Sonar for fast, lightweight answers and Sonar Pro for multi-step queries with more citations and a larger context window.",
      "highlights": [
        "API model IDs sonar and sonar-pro; Sonar Pro generally available to all developers at launch",
        "Model docs listed 127k context for sonar and 200k for sonar-pro, with an 8k output cap on sonar-pro",
        "Replaced the llama-3.1-sonar online models, which were retired on February 22, 2025",
        "JSON mode and search domain filters opened to select usage tiers"
      ],
      "source_title": "Introducing the Sonar Pro API by Perplexity",
      "source_url": "https://www.perplexity.ai/hub/blog/introducing-the-sonar-pro-api",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/01#perplexity-sonar-pro"
    },
    {
      "slug": "deepseek-r1",
      "date": "2025-01-20",
      "model": "DeepSeek-R1",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "MoE: 671B total, 37B active",
      "model_license": "MIT",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "DeepSeek-R1 is an open-weights reasoning model trained with large-scale reinforcement learning in post-training, released with a technical report. DeepSeek also released DeepSeek-R1-Zero, trained with reinforcement learning without supervised fine-tuning.",
      "highlights": [
        "Code and weights under the MIT License; API outputs may be used for fine-tuning and distillation",
        "Available in the API as deepseek-reasoner and as DeepThink on chat.deepseek.com",
        "128K context length per the model card"
      ],
      "source_title": "DeepSeek-R1 Release",
      "source_url": "https://api-docs.deepseek.com/news/news250120",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/01#deepseek-r1"
    },
    {
      "slug": "deepseek-r1-distill",
      "date": "2025-01-20",
      "model": "DeepSeek-R1-Distill",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "1.5B, 7B, 8B, 14B, 32B, 70B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "DeepSeek-R1-Distill is a set of six dense models fine-tuned on reasoning samples generated by DeepSeek-R1, based on Qwen2.5 and Llama 3 checkpoints. They bring R1-style reasoning to sizes from 1.5B to 70B parameters.",
      "highlights": [
        "Fine-tuned with 800K samples curated with DeepSeek-R1",
        "Qwen-based models derive from Qwen2.5 (Apache 2.0); Llama-based models from Llama 3.1 8B and Llama 3.3 70B",
        "Released the same day as DeepSeek-R1"
      ],
      "source_title": "DeepSeek-R1 Release",
      "source_url": "https://api-docs.deepseek.com/news/news250120",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/01#deepseek-r1-distill"
    },
    {
      "slug": "doubao-realtime-voice",
      "date": "2025-01-20",
      "model": "Doubao Realtime Voice Model",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "The Doubao Realtime Voice Model is an end-to-end speech understanding and generation model that replaces cascaded pipelines for spoken conversation, offering expressive voice control, emotional responsiveness, low latency, and smooth interruption.",
      "highlights": [
        "Fully available in the Doubao app from version 7.2.0",
        "Integrates speech and text modalities natively",
        "Supports natural interruptions during dialogue"
      ],
      "source_title": "Doubao Realtime Voice Model Is Available Upon Release! High EQ and IQ",
      "source_url": "https://seed.bytedance.com/en/blog/doubao-realtime-voice-model-is-available-upon-release-high-eq-and-iq",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/01#doubao-realtime-voice"
    },
    {
      "slug": "kimi-k1-5",
      "date": "2025-01-20",
      "model": "Kimi k1.5",
      "lab": "Moonshot AI",
      "lab_slug": "moonshot-ai",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Kimi k1.5 is Moonshot AI's multimodal reasoning model trained with reinforcement learning, which Moonshot describes as o1-level. Its training scaled the reinforcement learning context window to 128K and used partial rollouts for efficiency.",
      "highlights": [
        "Jointly trained on text and vision data",
        "Reports long chain-of-thought and short chain-of-thought results",
        "Moonshot said the service would come to kimi.ai soon after the January 20, 2025 report"
      ],
      "source_title": "Kimi k1.5: Scaling Reinforcement Learning with LLMs (GitHub)",
      "source_url": "https://github.com/MoonshotAI/Kimi-k1.5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/01#kimi-k1-5"
    },
    {
      "slug": "lfm-7b",
      "date": "2025-01-20",
      "model": "LFM-7B",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "7.7B",
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Cloud platforms",
        "Enterprise"
      ],
      "summary": "A 7.7B-parameter Liquid Foundation Model tuned for chat, instruction following, and multilingual use, with Liquid AI highlighting English, Arabic, and Japanese, aimed at private enterprise, on-premises, and on-device deployment.",
      "highlights": [
        "Supports English, Spanish, French, German, Chinese, Arabic, Japanese, and Korean",
        "Effective context length of 32k tokens validated with RULER",
        "Offered on Liquid Playground, OpenRouter, and AWS Marketplace",
        "Weights not released; on-premises customization stack sold to enterprises"
      ],
      "source_title": "Introducing LFM-7B: Setting New Standards for Efficient Language Models",
      "source_url": "https://www.liquid.ai/blog/introducing-lfm-7b-setting-new-standards-for-efficient-language-models",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/01#lfm-7b"
    },
    {
      "slug": "minimax-01",
      "date": "2025-01-15",
      "model": "MiniMax-Text-01 and MiniMax-VL-01",
      "lab": "MiniMax",
      "lab_slug": "minimax",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 4000000,
      "parameters": "MoE: 456B total, 45.9B active",
      "model_license": "MiniMax Model License",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "MiniMax-01 is an open-weight series pairing the MiniMax-Text-01 language model with the MiniMax-VL-01 vision-language model, built on a hybrid of Lightning Attention, softmax attention, and mixture-of-experts.",
      "highlights": [
        "456B total parameters with 45.9B activated per token",
        "Trained at a 1M-token context and handles up to 4M tokens at inference",
        "VL-01 adds a 303M-parameter vision transformer to the Text-01 base",
        "Model weights under the MiniMax Model License; code under MIT"
      ],
      "source_title": "MiniMax-01 is Now Open-Source: Scaling Lightning Attention for the AI Agent Era",
      "source_url": "https://www.minimax.io/news/minimax-01-series-2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/01#minimax-01"
    },
    {
      "slug": "luma-ray2",
      "date": "2025-01-15",
      "model": "Ray2",
      "lab": "Luma AI",
      "lab_slug": "luma-ai",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Second-generation Luma video model built on a new multimodal architecture trained with ten times the compute of Ray1, available in Dream Machine and later with learnable camera motion concepts.",
      "highlights": [
        "Ten times the compute of Ray1",
        "Camera Motion Concepts added March 31, 2025",
        "Announced for Adobe Firefly in April 2025"
      ],
      "source_title": "Changelog | Luma",
      "source_url": "https://lumalabs.ai/changelog",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/01#luma-ray2"
    },
    {
      "slug": "codestral-2501",
      "date": "2025-01-13",
      "model": "Codestral 25.01",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "coding",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Update of Codestral with a more efficient architecture and improved tokenizer that generates and completes code about twice as fast as the original, aimed at fill-in-the-middle use.",
      "highlights": [
        "Supports fill-in-the-middle, code correction and test generation",
        "Optimized for low-latency, high-frequency coding tasks",
        "API model ID codestral-2501"
      ],
      "source_title": "Codestral 25.01 | Mistral AI",
      "source_url": "https://mistral.ai/news/codestral-2501",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/01#codestral-2501"
    },
    {
      "slug": "minimax-i2v-01-live",
      "date": "2025-01-12",
      "model": "MiniMax I2V-01-live",
      "lab": "MiniMax",
      "lab_slug": "minimax",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "I2V-01-live is a Hailuo image-to-video model in the I2V family designed to animate 2D illustrations, with smoother, more vivid motion, greater stability, and subtle expressions across a range of art styles.",
      "highlights": [
        "Aimed at bringing 2D illustrations and characters to life",
        "Available in Hailuo AI and through the MiniMax API"
      ],
      "source_title": "I2V-01-live: smooth, vivid, and perfect for 2D illustrations",
      "source_url": "https://www.minimax.io/news/i2v-01-live-release",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/01#minimax-i2v-01-live"
    },
    {
      "slug": "minimax-s2v-01",
      "date": "2025-01-10",
      "model": "MiniMax S2V-01",
      "lab": "MiniMax",
      "lab_slug": "minimax",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "S2V-01 is Hailuo's subject reference video model, generating videos that keep a person's facial features and identity consistent across camera angles and movement from a single reference image.",
      "highlights": [
        "Needs only one reference image",
        "Posture, expression, and lighting can be adjusted with text prompts"
      ],
      "source_title": "S2V-01: Subject Reference of Hailuo",
      "source_url": "https://www.minimax.io/news/s2v-01-release",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/01#minimax-s2v-01"
    },
    {
      "slug": "cosmos-1-0",
      "date": "2025-01-06",
      "model": "Cosmos World Foundation Models",
      "lab": "NVIDIA",
      "lab_slug": "nvidia",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "NVIDIA Open Model License",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Platform of generative world foundation models for physical AI, with diffusion and autoregressive models, tokenizers and guardrails for robotics and autonomous vehicles.",
      "highlights": [
        "Announced at CES 2025",
        "Weights on Hugging Face and NVIDIA NGC under NVIDIA's open model license",
        "Previewable in the NVIDIA API catalog"
      ],
      "source_title": "NVIDIA Launches Cosmos World Foundation Model Platform to Accelerate Physical AI Development | NVIDIA Newsroom",
      "source_url": "https://nvidianews.nvidia.com/news/nvidia-launches-cosmos-world-foundation-model-platform-to-accelerate-physical-ai-development",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2025/01#cosmos-1-0"
    },
    {
      "slug": "deepseek-v3",
      "date": "2024-12-26",
      "model": "DeepSeek-V3",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "MoE: 671B total, 37B active",
      "model_license": "DeepSeek Model License",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "DeepSeek-V3 is an open-weights mixture-of-experts model built on Multi-head Latent Attention and the DeepSeekMoE architecture, pre-trained on 14.8T tokens. It replaced the model behind the deepseek-chat API and ran about 3 times faster than V2.",
      "highlights": [
        "671B total parameters with 37B activated per token",
        "Generates about 60 tokens per second, per DeepSeek",
        "Code under the MIT License; weights under the DeepSeek Model License, which permits commercial use"
      ],
      "source_title": "Introducing DeepSeek-V3",
      "source_url": "https://api-docs.deepseek.com/news/news1226",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/12#deepseek-v3"
    },
    {
      "slug": "qvq-72b-preview",
      "date": "2024-12-25",
      "model": "QVQ-72B-Preview",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "72B",
      "model_license": "Qwen License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "QVQ-72B-Preview is an experimental open-weights model for multimodal reasoning, built on Qwen2-VL-72B. It improves math-related visual reasoning compared with Qwen2-VL-72B-Instruct.",
      "highlights": [
        "Experimental research model with known limitations such as language mixing",
        "Built upon Qwen2-VL-72B",
        "Weights under the Qwen License Agreement"
      ],
      "source_title": "QVQ: To See the World with Wisdom",
      "source_url": "https://qwenlm.github.io/blog/qvq-72b-preview/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/12#qvq-72b-preview"
    },
    {
      "slug": "gemini-2-0-flash-thinking",
      "date": "2024-12-19",
      "model": "Gemini 2.0 Flash Thinking",
      "lab": "Google",
      "lab_slug": "google",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Gemini 2.0 Flash Thinking was an experimental test-time compute model that showed its thought process while generating a response, producing stronger reasoning than the base Gemini 2.0 Flash.",
      "highlights": [
        "Released for public preview in the Gemini API and Google AI Studio",
        "Updated as gemini-2.0-flash-thinking-exp-01-21 on Jan 21, 2025",
        "Added to the Gemini app model dropdown in February 2025"
      ],
      "source_title": "Release notes - Gemini API | Google AI Studio",
      "source_url": "https://ai.google.dev/gemini-api/docs/changelog",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/12#gemini-2-0-flash-thinking"
    },
    {
      "slug": "eleven-flash-v2-5",
      "date": "2024-12-18",
      "model": "Eleven Flash v2 and v2.5",
      "lab": "ElevenLabs",
      "lab_slug": "elevenlabs",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Ultra-low-latency text-to-speech models that generate speech in 75ms plus application and network latency, recommended by ElevenLabs for conversational voice agents and real-time use.",
      "highlights": [
        "Flash v2 is English only; Flash v2.5 supports 32 languages",
        "Model IDs eleven_flash_v2 and eleven_flash_v2_5",
        "Slightly lower quality than Turbo with lower latency"
      ],
      "source_title": "Meet Flash",
      "source_url": "https://elevenlabs.io/blog/meet-flash",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/12#eleven-flash-v2-5"
    },
    {
      "slug": "granite-3-1",
      "date": "2024-12-18",
      "model": "Granite 3.1",
      "lab": "IBM",
      "lab_slug": "ibm",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "2B and 8B dense; MoE 1B-A400M and 3B-A800M",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Update of the Granite 3 language models that extends every size to a 128K-token context window, released in base and instruct versions.",
      "highlights": [
        "128K-token context across dense and MoE models",
        "Tuned with synthetic data for long-context problems",
        "Released under Apache 2.0"
      ],
      "source_title": "ibm-granite/granite-3.1-8b-instruct · Hugging Face",
      "source_url": "https://huggingface.co/ibm-granite/granite-3.1-8b-instruct",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/12#granite-3-1"
    },
    {
      "slug": "granite-embedding",
      "date": "2024-12-18",
      "model": "Granite Embedding",
      "lab": "IBM",
      "lab_slug": "ibm",
      "type": "embedding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "embedding"
      ],
      "context_window": null,
      "parameters": "30M and 125M English; 107M and 278M multilingual",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "IBM's first Granite embedding models for retrieval, released as two English and two multilingual dense bi-encoder models under Apache 2.0.",
      "highlights": [
        "English models in 30M and 125M sizes",
        "Multilingual models in 107M and 278M sizes",
        "Followed by sparse and R2 versions in 2025"
      ],
      "source_title": "ibm-granite/granite-embedding-125m-english · Hugging Face",
      "source_url": "https://huggingface.co/ibm-granite/granite-embedding-125m-english",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/12#granite-embedding"
    },
    {
      "slug": "granite-guardian-3-1",
      "date": "2024-12-18",
      "model": "Granite Guardian 3.1",
      "lab": "IBM",
      "lab_slug": "ibm",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "2B and 8B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Updated Granite Guardian safety models built on Granite 3.1 for judging risks in prompts and responses, including harm and RAG hallucination checks.",
      "highlights": [
        "Released in 2B and 8B sizes",
        "Detects risks in prompts, responses and RAG outputs",
        "Released under Apache 2.0"
      ],
      "source_title": "ibm-granite/granite-guardian-3.1-8b · Hugging Face",
      "source_url": "https://huggingface.co/ibm-granite/granite-guardian-3.1-8b",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/12#granite-guardian-3-1"
    },
    {
      "slug": "leonardo-phoenix-1-0",
      "date": "2024-12-18",
      "model": "Phoenix 1.0",
      "lab": "Leonardo.Ai",
      "lab_slug": "leonardo-ai",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Full release of Leonardo.Ai's Phoenix foundational image model, with improved aesthetics over the preview, launched on the platform alongside Flux Dev and Flux Schnell.",
      "highlights": [
        "Offered in the API as phoenix-v1.0 alongside phoenix-v0.9",
        "Released on Replicate with Motion 2.0 on June 10, 2025"
      ],
      "source_title": "Leonardo.Ai on X: Flux + Phoenix 1.0 have landed.",
      "source_url": "https://x.com/LeonardoAi/status/1869351934576205869",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/12#leonardo-phoenix-1-0"
    },
    {
      "slug": "gpt-4o-mini-realtime-preview",
      "date": "2024-12-17",
      "model": "gpt-4o-mini-realtime-preview and gpt-4o-mini-audio-preview",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "OpenAI brought GPT-4o mini to the Realtime API beta and to audio in Chat Completions as preview speech models, alongside a new gpt-4o-realtime-preview-2024-12-17 snapshot and WebRTC support for the Realtime API.",
      "highlights": [
        "Realtime API model ID gpt-4o-mini-realtime-preview-2024-12-17",
        "Chat Completions model ID gpt-4o-mini-audio-preview-2024-12-17",
        "Announced with o1 in the API, Preference Fine-Tuning, and WebRTC support for the Realtime API",
        "Both snapshots are now listed as shut down on their API docs model pages"
      ],
      "source_title": "OpenAI o1 and new tools for developers",
      "source_url": "https://openai.com/index/o1-and-new-tools-for-developers/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/12#gpt-4o-mini-realtime-preview"
    },
    {
      "slug": "veo-2",
      "date": "2024-12-16",
      "model": "Veo 2",
      "lab": "Google",
      "lab_slug": "google",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API",
        "Cloud platforms"
      ],
      "summary": "Veo 2 is Google DeepMind's second-generation video generation model, producing clips at resolutions up to 4K that can extend to minutes in length. It rolled out first through Google Labs' VideoFX.",
      "highlights": [
        "Supports resolutions up to 4K",
        "Initial rollout in VideoFX through a waitlist",
        "Generally available in the Gemini API as veo-2.0-generate-001 from April 9, 2025"
      ],
      "source_title": "State-of-the-art video and image generation with Veo 2 and Imagen 3",
      "source_url": "https://blog.google/technology/google-labs/video-image-generation-update-december-2024/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/12#veo-2"
    },
    {
      "slug": "command-r7b",
      "date": "2024-12-13",
      "model": "Command R7B",
      "lab": "Cohere",
      "lab_slug": "cohere",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "7B",
      "model_license": "CC-BY-NC-4.0",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "The smallest, fastest and final model in Cohere's R series, built for RAG with citations, tool use and agents on modest hardware.",
      "highlights": [
        "128K-token context length",
        "Focus on math, reasoning, code and multilingual tasks",
        "Research weights released as c4ai-command-r7b-12-2024"
      ],
      "source_title": "Introducing Command R7B: Fast and efficient generative AI | Cohere Blog",
      "source_url": "https://cohere.com/blog/command-r7b",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/12#command-r7b"
    },
    {
      "slug": "deepseek-vl2",
      "date": "2024-12-13",
      "model": "DeepSeek-VL2",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "MoE: 1.0B, 2.8B, 4.5B active (Tiny, Small, VL2)",
      "model_license": "DeepSeek Model License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "DeepSeek-VL2 is a series of mixture-of-experts vision-language models that improves on DeepSeek-VL in visual question answering, OCR, document, table, and chart understanding, and visual grounding. It comes in Tiny, Small, and standard variants.",
      "highlights": [
        "Three variants with 1.0B, 2.8B, and 4.5B activated parameters",
        "Uses a dynamic tiling strategy for high-resolution images",
        "Model use is subject to the DeepSeek Model License, which supports commercial use"
      ],
      "source_title": "DeepSeek-VL2: Mixture-of-Experts Vision-Language Models (GitHub)",
      "source_url": "https://github.com/deepseek-ai/DeepSeek-VL2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/12#deepseek-vl2"
    },
    {
      "slug": "grok-2-1212",
      "date": "2024-12-12",
      "model": "Grok-2 (grok-2-1212) and Grok-2 Vision (grok-2-vision-1212)",
      "lab": "xAI",
      "lab_slug": "xai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "A new version of Grok-2, three times faster with improved accuracy, instruction following, and multilingual ability, rolled out free to all X users and was added to the API as grok-2-1212 and grok-2-vision-1212.",
      "highlights": [
        "API model IDs: grok-2-1212 and grok-2-vision-1212",
        "Free for all users on X, with higher limits for Premium and Premium+",
        "Docs release notes list API availability on December 14, 2024"
      ],
      "source_title": "Bringing Grok to Everyone | SpaceXSI",
      "source_url": "https://x.ai/news/grok-1212",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/12#grok-2-1212"
    },
    {
      "slug": "phi-4",
      "date": "2024-12-12",
      "model": "Phi-4",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 16000,
      "parameters": "14B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Phi-4 is a 14B Microsoft small language model specializing in complex reasoning, particularly math, with gains attributed to synthetic training data, curated organic data and improved post-training methods.",
      "highlights": [
        "14B parameter dense decoder-only model with a 16K token context length",
        "Announcement updated January 8, 2025, listing availability on Azure AI Foundry and Hugging Face",
        "Weights released under the MIT license"
      ],
      "source_title": "Introducing Phi-4: Microsoft's Newest Small Language Model Specializing in Complex Reasoning",
      "source_url": "https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/introducing-phi-4-microsoft%E2%80%99s-newest-small-language-model-specializing-in-comple/4357090",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/12#phi-4"
    },
    {
      "slug": "gemini-2-0-flash",
      "date": "2024-12-11",
      "model": "Gemini 2.0 Flash",
      "lab": "Google",
      "lab_slug": "google",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms",
        "Consumer app"
      ],
      "summary": "Gemini 2.0 Flash was the first model in the Gemini 2.0 family, a low-latency workhorse built for the agentic era with native tool use, released first as an experimental model alongside a new Multimodal Live API.",
      "highlights": [
        "Experimental release in the Gemini API, Google AI Studio and Vertex AI, plus a chat-optimized version in the Gemini app",
        "Native image generation and text-to-speech output were initially limited to early-access partners",
        "Generally available in the Gemini API and Vertex AI on Feb 5, 2025 with a 1 million token context window",
        "Shut down in the Gemini API on June 1, 2026"
      ],
      "source_title": "Google introduces Gemini 2.0: A new AI model for the agentic era",
      "source_url": "https://blog.google/innovation-and-ai/models-and-research/google-deepmind/google-gemini-ai-update-december-2024/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/12#gemini-2-0-flash"
    },
    {
      "slug": "deepseek-v2-5-1210",
      "date": "2024-12-10",
      "model": "DeepSeek-V2.5-1210",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "DeepSeek Model License",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "DeepSeek-V2.5-1210 is the final update in the DeepSeek-V2.5 series, improving math, coding, writing, and roleplay. DeepSeek released it alongside Internet Search on its web chat and described it as the close of the V2 line.",
      "highlights": [
        "Upgraded the deepseek-chat API model",
        "Optimized file upload and webpage summarization experiences",
        "Weights on Hugging Face as DeepSeek-V2.5-1210"
      ],
      "source_title": "DeepSeek V2.5: The Grand Finale",
      "source_url": "https://api-docs.deepseek.com/news/news1210",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/12#deepseek-v2-5-1210"
    },
    {
      "slug": "grok-aurora",
      "date": "2024-12-09",
      "model": "Aurora",
      "lab": "xAI",
      "lab_slug": "xai",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Aurora is xAI's image generation model for Grok, an autoregressive mixture-of-experts network trained to predict the next token from interleaved text and image data, with support for editing user-provided images.",
      "highlights": [
        "Launched on X in select countries, rolling out to all users within a week",
        "Trained on billions of internet examples, with a focus on photorealistic rendering",
        "Docs release notes list the image generation model on the API on March 19, 2025"
      ],
      "source_title": "Grok Image Generation Release | SpaceXSI",
      "source_url": "https://x.ai/news/grok-image-generation-release",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/12#grok-aurora"
    },
    {
      "slug": "sora",
      "date": "2024-12-09",
      "model": "Sora (Sora Turbo)",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Sora Turbo is a faster version of OpenAI's text-to-video model, released as a standalone product at sora.com for ChatGPT Plus and Pro users. It generates video from text, images, or existing video, moving Sora out of research preview.",
      "highlights": [
        "Generates videos up to 1080p and up to 20 seconds in widescreen, vertical, or square formats",
        "Significantly faster than the Sora model OpenAI previewed in February 2024",
        "Includes a storyboard tool; all generated videos carry C2PA metadata",
        "OpenAI notes the Sora product is no longer available as of April 26, 2026"
      ],
      "source_title": "Sora is here",
      "source_url": "https://openai.com/index/sora-is-here/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/12#sora"
    },
    {
      "slug": "llama-3-3",
      "date": "2024-12-06",
      "model": "Llama 3.3 70B",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "70B",
      "model_license": "Llama 3.3 Community License",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "Llama 3.3 is a 70B instruction-tuned multilingual text model with a 128K context window, released as a single model in the Llama 3 line.",
      "highlights": [
        "Officially supports eight languages, including English, German, Hindi and Thai",
        "Pretrained on about 15 trillion tokens with a December 2023 cutoff",
        "Text-only input; outputs text and code"
      ],
      "source_title": "meta-llama/Llama-3.3-70B-Instruct model card",
      "source_url": "https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/12#llama-3-3"
    },
    {
      "slug": "o1",
      "date": "2024-12-05",
      "model": "OpenAI o1 (with o1 pro mode)",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 200000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "OpenAI o1 is the next model in OpenAI's o1 reasoning series, succeeding o1-preview in ChatGPT. It launched alongside the ChatGPT Pro plan, which added o1 pro mode, a version of o1 that uses more compute to think harder.",
      "highlights": [
        "Reached the API on Dec 17, 2024 as o1-2024-12-17, rolling out to usage tier 5 developers",
        "API version supports function calling, Structured Outputs, developer messages, vision, and reasoning_effort",
        "200,000-token context window per the o1 API docs model page",
        "o1 pro mode is available to ChatGPT Pro subscribers"
      ],
      "source_title": "Introducing ChatGPT Pro",
      "source_url": "https://openai.com/index/introducing-chatgpt-pro/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/12#o1"
    },
    {
      "slug": "paligemma-2",
      "date": "2024-12-05",
      "model": "PaliGemma 2",
      "lab": "Google",
      "lab_slug": "google",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "3B, 10B, 28B",
      "model_license": "Gemma Terms of Use",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "PaliGemma 2 is Google's second-generation open vision-language model, built on Gemma 2 and offered in three sizes and three input resolutions. It adds detailed long captioning and is designed for easy fine-tuning.",
      "highlights": [
        "Input resolutions of 224px, 448px and 896px",
        "Drop-in replacement for existing PaliGemma users",
        "Pre-trained models on Hugging Face and Kaggle"
      ],
      "source_title": "Introducing PaliGemma 2: Powerful Vision-Language Models, Simple Fine-Tuning",
      "source_url": "https://developers.googleblog.com/en/introducing-paligemma-2-powerful-vision-language-models-simple-fine-tuning/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/12#paligemma-2"
    },
    {
      "slug": "snowflake-arctic-embed-2",
      "date": "2024-12-04",
      "model": "Arctic Embed 2.0",
      "lab": "Snowflake",
      "lab_slug": "snowflake",
      "type": "embedding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "embedding"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Multilingual second generation of Snowflake's Arctic Embed retrieval models, released as arctic-embed-l-v2.0 and arctic-embed-m-v2.0 for enterprise-grade multilingual search and retrieval.",
      "highlights": [
        "The large model has 303M non-embedding parameters",
        "Technical report released December 11, 2024",
        "Released under Apache 2.0"
      ],
      "source_title": "Snowflake/snowflake-arctic-embed-l-v2.0 · Hugging Face",
      "source_url": "https://huggingface.co/Snowflake/snowflake-arctic-embed-l-v2.0",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/12#snowflake-arctic-embed-2"
    },
    {
      "slug": "amazon-nova-canvas",
      "date": "2024-12-03",
      "model": "Amazon Nova Canvas",
      "lab": "Amazon",
      "lab_slug": "amazon",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Amazon Nova Canvas is Amazon's image generation model in the first Nova generation, creating images from text and image inputs with editing features such as inpainting, outpainting and background removal.",
      "highlights": [
        "Supports inpainting, outpainting and background removal",
        "Generated images include digital watermarking",
        "Virtual try-on and visual style options added July 1, 2025"
      ],
      "source_title": "Introducing Amazon Nova foundation models: Frontier intelligence and industry leading price performance",
      "source_url": "https://aws.amazon.com/blogs/aws/introducing-amazon-nova-frontier-intelligence-and-industry-leading-price-performance/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/12#amazon-nova-canvas"
    },
    {
      "slug": "amazon-nova-lite",
      "date": "2024-12-03",
      "model": "Amazon Nova Lite",
      "lab": "Amazon",
      "lab_slug": "amazon",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 300000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Amazon Nova Lite is a low-cost multimodal model in Amazon's first Nova generation that processes text, image and video input, with a context of up to 300K tokens and up to 30 minutes of video per request.",
      "highlights": [
        "Context up to 300K tokens",
        "Processes up to 30 minutes of video in a single request",
        "Launched in Amazon Bedrock"
      ],
      "source_title": "Introducing Amazon Nova foundation models: Frontier intelligence and industry leading price performance",
      "source_url": "https://aws.amazon.com/blogs/aws/introducing-amazon-nova-frontier-intelligence-and-industry-leading-price-performance/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/12#amazon-nova-lite"
    },
    {
      "slug": "amazon-nova-micro",
      "date": "2024-12-03",
      "model": "Amazon Nova Micro",
      "lab": "Amazon",
      "lab_slug": "amazon",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Amazon Nova Micro is the text-only model in Amazon's first Nova generation, focused on very fast, low-cost responses for tasks such as summarization, translation and classification, with a 128K token context window.",
      "highlights": [
        "Text-only model with a 128K token context window",
        "Launched in Amazon Bedrock with the rest of the Nova family",
        "Available in US East (N. Virginia), with more US Regions via cross-Region inference"
      ],
      "source_title": "Introducing Amazon Nova foundation models: Frontier intelligence and industry leading price performance",
      "source_url": "https://aws.amazon.com/blogs/aws/introducing-amazon-nova-frontier-intelligence-and-industry-leading-price-performance/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/12#amazon-nova-micro"
    },
    {
      "slug": "amazon-nova-pro",
      "date": "2024-12-03",
      "model": "Amazon Nova Pro",
      "lab": "Amazon",
      "lab_slug": "amazon",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 300000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Amazon Nova Pro is the multimodal model balancing accuracy, speed and cost in Amazon's first Nova generation, accepting text, image and video input with up to 300K input tokens.",
      "highlights": [
        "Accepts text, image and video input",
        "Up to 300K input tokens",
        "Launched in Amazon Bedrock"
      ],
      "source_title": "Introducing Amazon Nova foundation models: Frontier intelligence and industry leading price performance",
      "source_url": "https://aws.amazon.com/blogs/aws/introducing-amazon-nova-frontier-intelligence-and-industry-leading-price-performance/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/12#amazon-nova-pro"
    },
    {
      "slug": "amazon-nova-reel",
      "date": "2024-12-03",
      "model": "Amazon Nova Reel",
      "lab": "Amazon",
      "lab_slug": "amazon",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Amazon Nova Reel is Amazon's video generation model in the first Nova generation, producing short videos from text prompts and images with camera control actions such as zooming and camera movement.",
      "highlights": [
        "Supports camera control actions such as zooming or moving the camera",
        "Generated videos include digital watermarking",
        "Succeeded by Nova Reel 1.1 in April 2025"
      ],
      "source_title": "Introducing Amazon Nova foundation models: Frontier intelligence and industry leading price performance",
      "source_url": "https://aws.amazon.com/blogs/aws/introducing-amazon-nova-frontier-intelligence-and-industry-leading-price-performance/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/12#amazon-nova-reel"
    },
    {
      "slug": "hunyuanvideo",
      "date": "2024-12-03",
      "model": "HunyuanVideo",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Tencent Hunyuan Community License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "HunyuanVideo is Tencent's open text-to-video foundation model with over 13 billion parameters, using a multimodal LLM text encoder and joint image-video training, released with inference code and weights.",
      "highlights": [
        "Over 13 billion parameters",
        "Recommended output of 720x1280 at 129 frames",
        "FP8 weights added December 18, 2024"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/HunyuanVideo: HunyuanVideo: A Systematic Framework For Large Video Generation Model",
      "source_url": "https://github.com/Tencent-Hunyuan/HunyuanVideo",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/12#hunyuanvideo"
    },
    {
      "slug": "cohere-rerank-3-5",
      "date": "2024-12-02",
      "model": "Rerank 3.5",
      "lab": "Cohere",
      "lab_slug": "cohere",
      "type": "reranker",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "score"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms"
      ],
      "summary": "Reranking model with stronger reasoning over complex, constraint-heavy queries and broad data compatibility across long documents, semi-structured data and code.",
      "highlights": [
        "Covers 100+ languages with strength in 10 business languages",
        "Later made available on Azure AI Foundry",
        "API model ID rerank-v3.5"
      ],
      "source_title": "Introducing Rerank 3.5: Precise AI Search | Cohere Blog",
      "source_url": "https://cohere.com/blog/rerank-3pt5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/12#cohere-rerank-3-5"
    },
    {
      "slug": "qwq-32b-preview",
      "date": "2024-11-28",
      "model": "QwQ-32B-Preview",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32768,
      "parameters": "32B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "QwQ-32B-Preview (Qwen with Questions) is an experimental open-weights reasoning model from the Qwen team, focused on advancing AI reasoning in math, code, and general knowledge.",
      "highlights": [
        "Full 32,768-token context length",
        "Experimental release with limitations such as language mixing and recursive reasoning loops",
        "Weights under the Apache 2.0 license"
      ],
      "source_title": "QwQ: Reflect Deeply on the Boundaries of the Unknown",
      "source_url": "https://qwenlm.github.io/blog/qwq-32b-preview/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/11#qwq-32b-preview"
    },
    {
      "slug": "olmo-2",
      "date": "2024-11-26",
      "model": "OLMo 2",
      "lab": "Ai2",
      "lab_slug": "ai2",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "7B and 13B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Second generation of fully open OLMo models, trained on up to 5 trillion tokens and released with weights, data, code, recipes, checkpoints and instruct versions.",
      "highlights": [
        "Released in 7B and 13B sizes",
        "OLMo 2-Instruct models apply the Tülu 3 post-training recipe",
        "Released under Apache 2.0"
      ],
      "source_title": "OLMo 2: The best fully open language model to date | Ai2",
      "source_url": "https://allenai.org/blog/olmo2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/11#olmo-2"
    },
    {
      "slug": "smolvlm",
      "date": "2024-11-26",
      "model": "SmolVLM",
      "lab": "Hugging Face",
      "lab_slug": "hugging-face",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "2B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Compact 2B vision-language model built for small memory footprints, released with checkpoints, datasets, training recipes and tools under Apache 2.0.",
      "highlights": [
        "Released as SmolVLM-Base, SmolVLM-Synthetic and SmolVLM-Instruct",
        "Trained on the Cauldron and Docmatix datasets",
        "Integrated into transformers with a fine-tuning script"
      ],
      "source_title": "SmolVLM - small yet mighty Vision Language Model",
      "source_url": "https://huggingface.co/blog/smolvlm",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/11#smolvlm"
    },
    {
      "slug": "sd-3-5-large-controlnets",
      "date": "2024-11-26",
      "model": "Stable Diffusion 3.5 Large ControlNets",
      "lab": "Stability AI",
      "lab_slug": "stability-ai",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Stability AI Community License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Blur, Canny and Depth ControlNet models for Stable Diffusion 3.5 Large that add structural control to generations, including tiled upscaling to 8K and 16K resolution.",
      "highlights": [
        "Three control types: Blur, Canny and Depth",
        "Weights on Hugging Face with ComfyUI support",
        "Stability AI Community License"
      ],
      "source_title": "ControlNets for Stable Diffusion 3.5 Large",
      "source_url": "https://stability.ai/news-updates/sd3-5-large-controlnets",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/11#sd-3-5-large-controlnets"
    },
    {
      "slug": "runway-frames",
      "date": "2024-11-25",
      "model": "Frames",
      "lab": "Runway",
      "lab_slug": "runway",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Runway image generation base model focused on stylistic control and visual fidelity, able to hold a specific look across variations; it was later folded into Gen-4 Images.",
      "highlights": [
        "Became available to Unlimited and Enterprise plans on January 17, 2025",
        "Rolled out inside Gen-3 Alpha and the Runway API",
        "Frames is now Gen-4 Images"
      ],
      "source_title": "Introducing Frames",
      "source_url": "https://runway.com/research/introducing-frames",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/11#runway-frames"
    },
    {
      "slug": "flux-1-tools",
      "date": "2024-11-21",
      "model": "FLUX.1 Tools (Fill, Depth, Canny, Redux)",
      "lab": "Black Forest Labs",
      "lab_slug": "black-forest-labs",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "FLUX.1 [dev] Non-Commercial License",
      "available_at_launch": [
        "Open weights",
        "API",
        "Cloud platforms"
      ],
      "summary": "Suite of control and editing models for FLUX.1: Fill for inpainting and outpainting, Depth and Canny for structural guidance, and Redux for image variation, released as open [dev] weights and [pro] API endpoints.",
      "highlights": [
        "[dev] variants on Hugging Face under the FLUX.1 [dev] Non-Commercial License",
        "[pro] variants offered in the BFL API",
        "Depth and Canny were later deprecated in the BFL API; Fill and Redux remain"
      ],
      "source_title": "Introducing FLUX.1 Tools",
      "source_url": "https://bfl.ai/blog/24-11-21-tools",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/11#flux-1-tools"
    },
    {
      "slug": "ltx-video-0-9",
      "date": "2024-11-21",
      "model": "LTX-Video 0.9",
      "lab": "Lightricks",
      "lab_slug": "lightricks",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": "2B",
      "model_license": "RAIL-M",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Initial open release of Lightricks' LTX-Video, a diffusion transformer video model supporting text-to-video and image-to-video generation, with a research paper following in December 2024.",
      "highlights": [
        "Text-to-video and image-to-video",
        "Version 0.9.1 checkpoint with improved quality released December 20, 2024",
        "Weights on Hugging Face"
      ],
      "source_title": "Lightricks/LTX-Video",
      "source_url": "https://github.com/Lightricks/LTX-Video",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/11#ltx-video-0-9"
    },
    {
      "slug": "tulu-3",
      "date": "2024-11-21",
      "model": "Tülu 3",
      "lab": "Ai2",
      "lab_slug": "ai2",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "8B and 70B",
      "model_license": "Llama 3.1 Community License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Family of post-trained Llama 3.1 models released with all of their data, data mixes, recipes, code, infrastructure and evaluation framework.",
      "highlights": [
        "Opens up the full post-training recipe",
        "Built on Llama 3.1 base models",
        "Later scaled to 405B parameters"
      ],
      "source_title": "Tülu 3 opens language model post-training up to more tasks and more people | Ai2",
      "source_url": "https://allenai.org/blog/tulu-3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/11#tulu-3"
    },
    {
      "slug": "deepseek-r1-lite-preview",
      "date": "2024-11-20",
      "model": "DeepSeek-R1-Lite-Preview",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "DeepSeek-R1-Lite-Preview is a reasoning model preview that shows its thought process in real time. It launched on chat.deepseek.com, with DeepSeek saying open-source models and an API would follow.",
      "highlights": [
        "Available on DeepSeek's web chat at launch, without API access",
        "Showed steadily improving results as thought length increased, per DeepSeek",
        "Preceded the open-weights DeepSeek-R1 release in January 2025"
      ],
      "source_title": "DeepSeek-R1-Lite-Preview is now live: unleashing supercharged reasoning power!",
      "source_url": "https://api-docs.deepseek.com/news/news1120",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/11#deepseek-r1-lite-preview"
    },
    {
      "slug": "suno-v4",
      "date": "2024-11-19",
      "model": "Suno v4",
      "lab": "Suno",
      "lab_slug": "suno",
      "type": "music",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Music model refining v3 with higher audio quality, launched with Remaster for upgrading older tracks, the ReMi lyrics model and v4-powered Covers and Personas.",
      "highlights": [
        "Remaster upgrades older songs to v4 quality",
        "New ReMi lyrics assist model",
        "Powers Covers and Personas"
      ],
      "source_title": "Introducing v4",
      "source_url": "https://suno.com/blog/v4",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/11#suno-v4"
    },
    {
      "slug": "mistral-large-2411",
      "date": "2024-11-18",
      "model": "Mistral Large 24.11",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "123B",
      "model_license": "Mistral Research License",
      "available_at_launch": [
        "Open weights",
        "API",
        "Cloud platforms"
      ],
      "summary": "Update of Mistral Large 2 with better long-context understanding, a new system prompt and more accurate function calling, released with weights under the Mistral Research License.",
      "highlights": [
        "Also called Mistral Large 2.1 in Mistral's changelog",
        "API model ID mistral-large-2411",
        "Announced for Google Cloud and Microsoft Azure"
      ],
      "source_title": "Pixtral Large | Mistral AI",
      "source_url": "https://mistral.ai/news/pixtral-large",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/11#mistral-large-2411"
    },
    {
      "slug": "pixtral-large",
      "date": "2024-11-18",
      "model": "Pixtral Large",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "124B: 123B decoder + 1B vision encoder",
      "model_license": "Mistral Research License",
      "available_at_launch": [
        "Open weights",
        "API",
        "Consumer app"
      ],
      "summary": "Frontier-class multimodal model that extends Mistral Large 2 with a 1B-parameter vision encoder, powering document and image understanding in Le Chat.",
      "highlights": [
        "128K-token context fits at least 30 high-resolution images",
        "API model ID pixtral-large-latest",
        "Weights released under the Mistral Research License"
      ],
      "source_title": "Pixtral Large | Mistral AI",
      "source_url": "https://mistral.ai/news/pixtral-large",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/11#pixtral-large"
    },
    {
      "slug": "qwen2-5-turbo",
      "date": "2024-11-15",
      "model": "Qwen2.5-Turbo",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Qwen2.5-Turbo is a version of Qwen2.5 that supports a 1M-token context, offered through the Alibaba Cloud Model Studio API. It uses sparse attention mechanisms to speed up long-context processing.",
      "highlights": [
        "Time to first token for 1M-token context cut from 4.9 minutes to 68 seconds",
        "Compatible with the OpenAI API format",
        "Demos available on Hugging Face and ModelScope"
      ],
      "source_title": "Extending the Context Length to 1M Tokens!",
      "source_url": "https://qwenlm.github.io/blog/qwen2.5-turbo/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/11#qwen2-5-turbo"
    },
    {
      "slug": "janusflow",
      "date": "2024-11-13",
      "model": "JanusFlow",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text",
        "image"
      ],
      "context_window": null,
      "parameters": "1.3B",
      "model_license": "DeepSeek Model License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "JanusFlow is a unified model for image understanding and generation that integrates an autoregressive language model with rectified flow. DeepSeek reports that rectified flow can be trained within the language model framework without complex architectural changes.",
      "highlights": [
        "Released as JanusFlow-1.3B on Hugging Face",
        "Part of DeepSeek's Janus series of unified multimodal models",
        "Model use under the DeepSeek Model License"
      ],
      "source_title": "Janus-Series: Unified Multimodal Understanding and Generation Models (GitHub)",
      "source_url": "https://github.com/deepseek-ai/Janus",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/11#janusflow"
    },
    {
      "slug": "qwen2-5-coder-32b",
      "date": "2024-11-12",
      "model": "Qwen2.5-Coder (0.5B to 32B)",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 131072,
      "parameters": "0.5B, 1.5B, 3B, 7B, 14B, 32B",
      "model_license": "Apache-2.0 (3B uses the Qwen Research License Agreement)",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "The Qwen2.5-Coder series expanded to six sizes with the release of 0.5B, 3B, 14B, and 32B models, led by Qwen2.5-Coder-32B-Instruct. Qwen says the 32B model matches GPT-4o's coding capabilities.",
      "highlights": [
        "Adds four sizes to the earlier 1.5B and 7B models",
        "Full 131,072-token context for the larger models",
        "Most sizes under Apache 2.0"
      ],
      "source_title": "Qwen2.5-Coder Series: Powerful, Diverse, Practical.",
      "source_url": "https://qwenlm.github.io/blog/qwen2.5-coder-family/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/11#qwen2-5-coder-32b"
    },
    {
      "slug": "cogvideox-1-5",
      "date": "2024-11-08",
      "model": "CogVideoX1.5",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": "5B",
      "model_license": "CogVideoX License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "CogVideoX1.5 is an upgraded version of Z.ai's open video generation model CogVideoX. The CogVideoX1.5-5B series generates 10-second videos at higher resolution, and its image-to-video variant supports any resolution.",
      "highlights": [
        "Released as CogVideoX1.5-5B and CogVideoX1.5-5B-I2V",
        "Diffusers version followed on November 15, 2024",
        "Weights under the CogVideoX License"
      ],
      "source_title": "CogVideo & CogVideoX (GitHub)",
      "source_url": "https://github.com/zai-org/CogVideo",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/11#cogvideox-1-5"
    },
    {
      "slug": "mistral-moderation",
      "date": "2024-11-07",
      "model": "Mistral Moderation",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 8000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "LLM-based content moderation classifier that sorts text into nine policy categories, offered as an API with endpoints for raw text and conversational content.",
      "highlights": [
        "The same model powers moderation in Le Chat",
        "Covers categories including unqualified advice and PII",
        "API model ID mistral-moderation-2411"
      ],
      "source_title": "Mistral Moderation API | Mistral AI",
      "source_url": "https://mistral.ai/news/mistral-moderation",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/11#mistral-moderation"
    },
    {
      "slug": "flux-1-1-pro-ultra-raw",
      "date": "2024-11-06",
      "model": "FLUX1.1 [pro] Ultra and Raw modes",
      "lab": "Black Forest Labs",
      "lab_slug": "black-forest-labs",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "High-resolution and photorealism modes for FLUX1.1 [pro]: Ultra generates images at four times standard resolution, up to 4MP, while Raw produces a less synthetic, candid-photography look.",
      "highlights": [
        "Ultra mode supports up to 4MP output",
        "Raw mode increases diversity in human subjects",
        "Available through the BFL API"
      ],
      "source_title": "Introducing FLUX1.1 [pro] Ultra and Raw Modes",
      "source_url": "https://bfl.ai/blog/24-11-06-ultra",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/11#flux-1-1-pro-ultra-raw"
    },
    {
      "slug": "hunyuan-large",
      "date": "2024-11-05",
      "model": "Hunyuan-Large",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "MoE: 389B total, 52B active",
      "model_license": "Tencent Hunyuan Community License Agreement",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "Hunyuan-Large (Hunyuan-MoE-A52B) is Tencent's open Transformer-based mixture-of-experts language model with 389B total and 52B active parameters, released with a technical report and training and inference tooling.",
      "highlights": [
        "Pretrained model supports 256K tokens; the Instruct model supports 128K",
        "Pretrain, Instruct, and Instruct-FP8 checkpoints on Hugging Face",
        "Integrated into Tencent Cloud's TI platform at launch"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/Tencent-Hunyuan-Large",
      "source_url": "https://github.com/Tencent-Hunyuan/Tencent-Hunyuan-Large",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/11#hunyuan-large"
    },
    {
      "slug": "hunyuan3d-1-0",
      "date": "2024-11-05",
      "model": "Hunyuan3D-1.0",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "3d"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Tencent Hunyuan Non-Commercial License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Hunyuan3D-1.0 is a two-stage open framework for text-to-3D and image-to-3D generation that combines fast multi-view diffusion with feed-forward reconstruction, released in lite and standard versions.",
      "highlights": [
        "Lite model produces a mesh in about 10 seconds and the standard model in about 25 seconds",
        "Uses Hunyuan-DiT for text-conditioned generation"
      ],
      "source_title": "GitHub - Tencent-Hunyuan/Hunyuan3D-1: Tencent Hunyuan3D-1.0: A Unified Framework for Text-to-3D and Image-to-3D Generation",
      "source_url": "https://github.com/Tencent-Hunyuan/Hunyuan3D-1",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/11#hunyuan3d-1-0"
    },
    {
      "slug": "grok-beta",
      "date": "2024-11-04",
      "model": "grok-beta",
      "lab": "xAI",
      "lab_slug": "xai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "grok-beta was the preview model that opened xAI's API public beta, a provisional name for a new Grok model then in the final stages of development, supporting function calling and system prompts.",
      "highlights": [
        "128,000-token context window",
        "REST API compatible with OpenAI and Anthropic SDKs",
        "An image-input preview, grok-vision-beta, followed"
      ],
      "source_title": "API Public Beta | SpaceXSI",
      "source_url": "https://x.ai/news/api",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/11#grok-beta"
    },
    {
      "slug": "recraft-v3",
      "date": "2024-10-30",
      "model": "Recraft V3",
      "lab": "Recraft",
      "lab_slug": "recraft",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Image model trained from scratch with major gains in text generation, able to render long text in images, plus anatomical accuracy, prompt understanding and vector image generation.",
      "highlights": [
        "Generates long texts, not just a few words, within images",
        "Supports vector art generation and style consistency",
        "Launched with a Recraft API for developers"
      ],
      "source_title": "Introducing Recraft V3: #1 AI Image Generation Model",
      "source_url": "https://www.recraft.ai/blog/recraft-introduces-a-revolutionary-ai-model-that-thinks-in-design-language",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/10#recraft-v3"
    },
    {
      "slug": "stable-diffusion-3-5-medium",
      "date": "2024-10-29",
      "model": "Stable Diffusion 3.5 Medium",
      "lab": "Stability AI",
      "lab_slug": "stability-ai",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "2.5B",
      "model_license": "Stability AI Community License",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "2.5B-parameter Stable Diffusion 3.5 model with an improved MMDiT-X architecture, designed to run on standard consumer hardware and generate images between 0.25 and 2 megapixels.",
      "highlights": [
        "2.5B parameters",
        "Requires about 9.9 GB of VRAM excluding text encoders",
        "Released one week after SD 3.5 Large"
      ],
      "source_title": "Introducing Stable Diffusion 3.5",
      "source_url": "https://stability.ai/news-updates/introducing-stable-diffusion-3-5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/10#stable-diffusion-3-5-medium"
    },
    {
      "slug": "aya-expanse",
      "date": "2024-10-24",
      "model": "Aya Expanse",
      "lab": "Cohere",
      "lab_slug": "cohere",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "8B and 32B",
      "model_license": "CC-BY-NC-4.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Family of multilingual open-weights models from Cohere For AI covering 23 languages, built by combining data arbitrage, multilingual preference training and model merging.",
      "highlights": [
        "Released on Kaggle and Hugging Face",
        "Made available to chat with on WhatsApp in January 2025",
        "Weights published under a non-commercial license"
      ],
      "source_title": "Cohere For AI, Cohere's research arm, announces Aya Expanse",
      "source_url": "https://cohere.com/blog/aya-expanse-connecting-our-world",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/10#aya-expanse"
    },
    {
      "slug": "glm-4-voice",
      "date": "2024-10-24",
      "model": "GLM-4-Voice",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": null,
      "parameters": "9B",
      "model_license": "GLM-4-Voice License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "GLM-4-Voice is Z.ai's end-to-end voice model that understands and generates Chinese and English speech for real-time voice conversation. Users can change emotion, intonation, speech rate, and dialect by instruction.",
      "highlights": [
        "Speech tokenizer converts each second of audio into 12.5 discrete tokens",
        "GLM-4-Voice-9B is pre-trained on speech based on GLM-4-9B",
        "Streaming decoder can start generating after as few as 10 audio tokens"
      ],
      "source_title": "GLM-4-Voice (GitHub)",
      "source_url": "https://github.com/zai-org/GLM-4-Voice",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/10#glm-4-voice"
    },
    {
      "slug": "llama-3-2-quantized",
      "date": "2024-10-24",
      "model": "Llama 3.2 1B and 3B quantized",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "1B, 3B",
      "model_license": "Llama 3.2 Community License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Meta released quantized versions of Llama 3.2 1B and 3B, its first lightweight quantized Llama models, built with QAT plus LoRA and with SpinQuant to run on mobile devices.",
      "highlights": [
        "2 to 4 times faster than the original BF16 models on average",
        "Average 56% smaller model size and 41% lower memory use",
        "Model card lists an 8K context length for the quantized models"
      ],
      "source_title": "Introducing quantized Llama models with increased speed and a reduced memory footprint",
      "source_url": "https://ai.meta.com/blog/meta-llama-quantized-lightweight-models/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/10#llama-3-2-quantized"
    },
    {
      "slug": "runway-act-one",
      "date": "2024-10-22",
      "model": "Act-One",
      "lab": "Runway",
      "lab_slug": "runway",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "video",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Tool inside Gen-3 Alpha for generating expressive character performances from a single driving video of an actor and a character image, without motion capture or rigging.",
      "highlights": [
        "Captures eye-lines, micro expressions, pacing and delivery",
        "Rolled out gradually to users from October 2024",
        "Succeeded by Act-Two in July 2025"
      ],
      "source_title": "Introducing Act-One",
      "source_url": "https://runway.com/research/introducing-act-one",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/10#runway-act-one"
    },
    {
      "slug": "claude-3-5-sonnet-2024-10",
      "date": "2024-10-22",
      "model": "Claude 3.5 Sonnet (upgraded) and Claude 3.5 Haiku",
      "lab": "Anthropic",
      "lab_slug": "anthropic",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Cloud platforms"
      ],
      "summary": "Anthropic released an upgraded Claude 3.5 Sonnet with gains in agentic coding and tool use, plus the new Claude 3.5 Haiku. Claude 3.5 Sonnet was the first Anthropic model to offer computer use, in public beta.",
      "highlights": [
        "API model IDs claude-3-5-sonnet-20241022 and claude-3-5-haiku-20241022",
        "Computer use public beta on the Anthropic API, Amazon Bedrock, and Google Cloud Vertex AI",
        "Claude 3.5 Haiku reached the Claude API on November 4, 2024 as a text-only model",
        "Upgraded Claude 3.5 Sonnet available to all Claude users at launch"
      ],
      "source_title": "Introducing computer use, a new Claude 3.5 Sonnet, and Claude 3.5 Haiku",
      "source_url": "https://www.anthropic.com/news/3-5-models-and-computer-use",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/10#claude-3-5-sonnet-2024-10"
    },
    {
      "slug": "cohere-embed-3-multimodal",
      "date": "2024-10-22",
      "model": "Embed 3 (multimodal)",
      "lab": "Cohere",
      "lab_slug": "cohere",
      "type": "embedding",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "embedding"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Update that makes Cohere's Embed 3 models multimodal, generating embeddings from both text and images in the same vector space for enterprise search.",
      "highlights": [
        "Text and image embeddings share one vector space",
        "Targets charts, product catalogs and design files",
        "Later offered on Amazon Bedrock"
      ],
      "source_title": "Introducing multimodal Embed 3: Powering AI search | Cohere Blog",
      "source_url": "https://cohere.com/blog/multimodal-embed-3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/10#cohere-embed-3-multimodal"
    },
    {
      "slug": "genmo-mochi-1",
      "date": "2024-10-22",
      "model": "Mochi 1 (preview)",
      "lab": "Genmo",
      "lab_slug": "genmo",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": "10B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "Consumer app"
      ],
      "summary": "Research preview of Genmo's open text-to-video model, a 10B-parameter diffusion model built on an Asymmetric Diffusion Transformer architecture with strong motion quality and prompt adherence.",
      "highlights": [
        "10B parameters, trained from scratch",
        "Generates 480p video at 30 fps for up to 5.4 seconds",
        "Hosted playground launched alongside the weights"
      ],
      "source_title": "Mochi 1: A new SOTA in open text-to-video",
      "source_url": "https://www.genmo.ai/blog/mochi-1-a-new-sota-in-open-text-to-video",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/10#genmo-mochi-1"
    },
    {
      "slug": "stable-diffusion-3-5-large",
      "date": "2024-10-22",
      "model": "Stable Diffusion 3.5 Large and Large Turbo",
      "lab": "Stability AI",
      "lab_slug": "stability-ai",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "8.1B",
      "model_license": "Stability AI Community License",
      "available_at_launch": [
        "Open weights",
        "API",
        "Cloud platforms"
      ],
      "summary": "Open Stable Diffusion 3.5 release led by the 8.1B-parameter Large base model for professional 1MP work and a distilled Large Turbo variant that generates images in four steps.",
      "highlights": [
        "Stable Diffusion 3.5 Large has 8.1B parameters",
        "Large Turbo generates in 4 steps",
        "Free commercial use for smaller organizations under the Stability AI Community License"
      ],
      "source_title": "Introducing Stable Diffusion 3.5",
      "source_url": "https://stability.ai/news-updates/introducing-stable-diffusion-3-5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/10#stable-diffusion-3-5-large"
    },
    {
      "slug": "granite-3-0",
      "date": "2024-10-21",
      "model": "Granite 3.0",
      "lab": "IBM",
      "lab_slug": "ibm",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 4096,
      "parameters": "2B and 8B dense; MoE 1B-A400M and 3B-A800M",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Third generation of IBM's open language models, released as dense 2B and 8B models and sparse mixture-of-experts 1B and 3B models in base and instruct versions.",
      "highlights": [
        "4,096-token sequence length",
        "MoE models activate 400M and 800M parameters",
        "Released under Apache 2.0"
      ],
      "source_title": "ibm-granite/granite-3.0-8b-instruct · Hugging Face",
      "source_url": "https://huggingface.co/ibm-granite/granite-3.0-8b-instruct",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/10#granite-3-0"
    },
    {
      "slug": "granite-guardian-3-0",
      "date": "2024-10-21",
      "model": "Granite Guardian 3.0",
      "lab": "IBM",
      "lab_slug": "ibm",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "2B and 8B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Safety models fine-tuned from Granite 3.0 Instruct to detect risks in prompts and responses, including harm, jailbreaking and RAG issues such as groundedness and context relevance.",
      "highlights": [
        "Released in 2B and 8B sizes",
        "Covers context relevance, groundedness and answer relevance for RAG",
        "Released under Apache 2.0"
      ],
      "source_title": "ibm-granite/granite-guardian-3.0-8b · Hugging Face",
      "source_url": "https://huggingface.co/ibm-granite/granite-guardian-3.0-8b",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/10#granite-guardian-3-0"
    },
    {
      "slug": "janus",
      "date": "2024-10-18",
      "model": "Janus",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text",
        "image"
      ],
      "context_window": null,
      "parameters": "1.3B",
      "model_license": "DeepSeek Model License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Janus is an autoregressive framework that unifies multimodal understanding and image generation by decoupling visual encoding into separate pathways while using a single transformer. DeepSeek released it as Janus-1.3B.",
      "highlights": [
        "Separate visual encoding paths for understanding and generation",
        "Single unified transformer architecture",
        "Code under the MIT License; model use under the DeepSeek Model License"
      ],
      "source_title": "deepseek-ai/Janus-1.3B (Hugging Face)",
      "source_url": "https://huggingface.co/deepseek-ai/Janus-1.3B",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/10#janus"
    },
    {
      "slug": "ministral-3b",
      "date": "2024-10-16",
      "model": "Ministral 3B",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "small",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "3B",
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Compact 3B edge model introduced as one of les Ministraux for local, privacy-first and low-latency applications such as on-device translation and offline assistants.",
      "highlights": [
        "Supports up to 128K tokens of context",
        "Offered through the API rather than as open weights",
        "API model ID ministral-3b-2410"
      ],
      "source_title": "Un Ministral, des Ministraux | Mistral AI",
      "source_url": "https://mistral.ai/news/ministraux",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/10#ministral-3b"
    },
    {
      "slug": "ministral-8b",
      "date": "2024-10-16",
      "model": "Ministral 8B",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "8B",
      "model_license": "Mistral Research License",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Edge-focused 8B model introduced with Ministral 3B for on-device and low-latency use, featuring an interleaved sliding-window attention pattern for memory-efficient inference.",
      "highlights": [
        "Supports up to 128K tokens of context",
        "Suited to function calling in multi-step agent workflows",
        "API model ID ministral-8b-2410"
      ],
      "source_title": "Un Ministral, des Ministraux | Mistral AI",
      "source_url": "https://mistral.ai/news/ministraux",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/10#ministral-8b"
    },
    {
      "slug": "inflection-3",
      "date": "2024-10-07",
      "model": "Inflection 3.0 (Pi 3.0 and Productivity 3.0)",
      "lab": "Inflection AI",
      "lab_slug": "inflection-ai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Enterprise"
      ],
      "summary": "Inflection AI's enterprise-grade model generation, launched with Inflection for Enterprise and a commercial API, comprising Pi 3.0 for emotionally intelligent conversation and Productivity 3.0 for instruction following and JSON output.",
      "highlights": [
        "API config IDs inflection_3_pi and inflection_3_productivity at developers.inflection.ai",
        "Both models supported an 8k context window, per the developer docs",
        "Pi 3.0 was described as the model powering the Pi experience",
        "Runs on Intel Gaudi 3 accelerators and Intel Tiber AI Cloud"
      ],
      "source_title": "Introducing Inflection for Enterprise",
      "source_url": "https://inflection.ai/blog/enterprise",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/10#inflection-3"
    },
    {
      "slug": "reka-flash-2024-10",
      "date": "2024-10-04",
      "model": "Reka Flash (October 2024 update)",
      "lab": "Reka",
      "lab_slug": "reka",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video",
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "21B",
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Enterprise"
      ],
      "summary": "An updated Reka Flash with function calling, structured output, stronger instruction following, native audio understanding in video, and support for 3 to 5 minute videos, up from about one minute in the first version.",
      "highlights": [
        "Interleaved text, image, video and audio input with a 128K-token context length",
        "Available on Reka Chat and the Reka API, plus on-premises and private cloud deployment",
        "Speech output was experimental and not in the public API at launch"
      ],
      "source_title": "Reka Flash Updates: Advanced Multimodal Understanding, Improved Reasoning, Better Agent Building Blocks and more.",
      "source_url": "https://reka.ai/news/reka-flash-updates",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/10#reka-flash-2024-10"
    },
    {
      "slug": "flux-1-1-pro",
      "date": "2024-10-02",
      "model": "FLUX1.1 [pro]",
      "lab": "Black Forest Labs",
      "lab_slug": "black-forest-labs",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms"
      ],
      "summary": "Successor to FLUX.1 [pro] with faster generation and improved image quality, prompt adherence and diversity, launched together with general availability of the beta BFL API.",
      "highlights": [
        "Tested under the codename blueberry before launch",
        "Released with the beta BFL API",
        "Also offered through Together.ai, Replicate, fal.ai and Freepik"
      ],
      "source_title": "Announcing FLUX1.1 [pro] and the BFL API",
      "source_url": "https://bfl.ai/blog/24-10-02-flux",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/10#flux-1-1-pro"
    },
    {
      "slug": "gpt-4o-realtime-preview",
      "date": "2024-10-01",
      "model": "gpt-4o-realtime-preview and gpt-4o-audio-preview",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "OpenAI launched the Realtime API in public beta with gpt-4o-realtime-preview, a GPT-4o model for low-latency speech-to-speech over WebSockets, and announced gpt-4o-audio-preview to bring audio input and output to the Chat Completions API.",
      "highlights": [
        "Realtime API public beta opened to all paid developers, with six preset voices and function calling",
        "gpt-4o-audio-preview released in Chat Completions on Oct 17, 2024 (API changelog)",
        "Five more voices added Oct 30, 2024; updated 2024-12-17 snapshots followed",
        "OpenAI announced Realtime API general availability on Aug 28, 2025"
      ],
      "source_title": "Introducing the Realtime API",
      "source_url": "https://openai.com/index/introducing-the-realtime-api/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/10#gpt-4o-realtime-preview"
    },
    {
      "slug": "whisper-large-v3-turbo",
      "date": "2024-10-01",
      "model": "Whisper large-v3-turbo",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "809M",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Whisper large-v3-turbo is an open-weights speech recognition model that fine-tunes a pruned Whisper large-v3, cutting decoder layers from 32 to 4 for faster transcription with minimal accuracy degradation.",
      "highlights": [
        "Called turbo in the whisper package; weights on Hugging Face as openai/whisper-large-v3-turbo",
        "809M parameters versus 1550M for large-v3",
        "Not trained for translation tasks; it returns the original language",
        "Code and weights shipped in the whisper v20240930 GitHub release"
      ],
      "source_title": "turbo model release",
      "source_url": "https://github.com/openai/whisper/discussions/2363",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/10#whisper-large-v3-turbo"
    },
    {
      "slug": "lfm-1b-3b-40b",
      "date": "2024-09-30",
      "model": "LFM-1B, LFM-3B, and LFM-40B",
      "lab": "Liquid AI",
      "lab_slug": "liquid-ai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "1B: 1.3B; 3B: 3.1B; 40B: 40.3B MoE (12B active)",
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "Enterprise"
      ],
      "summary": "Liquid AI's first series of Liquid Foundation Models, non-transformer language models built from dynamical systems and signal processing principles, released as dense 1.3B and 3.1B models plus a 40.3B mixture-of-experts model.",
      "highlights": [
        "Preview release optimized for a 32k-token context length",
        "LFM-40B activates 12B parameters per token",
        "Available on Liquid Playground; weights not open-sourced",
        "Liquid AI now marks this generation as deprecated"
      ],
      "source_title": "Liquid Foundation Models: Our First Series of Generative AI Models",
      "source_url": "https://www.liquid.ai/blog/liquid-foundation-models-our-first-series-of-generative-ai-models",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/09#lfm-1b-3b-40b"
    },
    {
      "slug": "meta-sam-2-1",
      "date": "2024-09-30",
      "model": "SAM 2.1",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "image",
        "video"
      ],
      "output_modalities": [
        "image",
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "SAM 2.1 is an updated set of Segment Anything Model 2 checkpoints with stronger performance, released with the SAM 2 training code and web demo code as the SAM 2.1 Developer Suite.",
      "highlights": [
        "New checkpoints require the latest SAM 2 repository code",
        "Training and fine-tuning code released",
        "Front-end and back-end code for the SAM 2 web demo released"
      ],
      "source_title": "facebookresearch/sam2 GitHub repository",
      "source_url": "https://github.com/facebookresearch/sam2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/09#meta-sam-2-1"
    },
    {
      "slug": "cogview3-plus",
      "date": "2024-09-29",
      "model": "CogView3-Plus-3B",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "3B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "CogView-3Plus-3B is Z.ai's open-weights text-to-image model, released together with the CogView3 text-to-image system. Z.ai added a diffusers version of the model on October 13, 2024.",
      "highlights": [
        "Released alongside CogView3 on September 29, 2024",
        "Diffusers support added October 13, 2024",
        "Weights under the Apache 2.0 license"
      ],
      "source_title": "CogView4 & CogView3 & CogView-3Plus (GitHub)",
      "source_url": "https://github.com/zai-org/CogView4",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/09#cogview3-plus"
    },
    {
      "slug": "omni-moderation-latest",
      "date": "2024-09-26",
      "model": "omni-moderation-latest",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "omni-moderation-latest is a GPT-4o-based model for OpenAI's free Moderation API that classifies both text and images, adds two text-only harm categories, and improves accuracy, especially for non-English content.",
      "highlights": [
        "Snapshot ID omni-moderation-2024-09-26",
        "Image inputs supported for violence, self-harm, and sexual categories",
        "Adds illicit and illicit/violent text categories",
        "Free for all developers, with rate limits by usage tier"
      ],
      "source_title": "Upgrading the Moderation API with our new multimodal moderation model",
      "source_url": "https://openai.com/index/upgrading-the-moderation-api-with-our-new-multimodal-moderation-model/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/09#omni-moderation-latest"
    },
    {
      "slug": "llama-3-2-vision",
      "date": "2024-09-25",
      "model": "Llama 3.2 11B and 90B Vision",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "11B, 90B",
      "model_license": "Llama 3.2 Community License",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "Llama 3.2 Vision adds image understanding to Llama with 11B and 90B parameter models that take text and images as input and produce text output.",
      "highlights": [
        "Model card lists a 128K context length",
        "English is the only supported language for image plus text tasks",
        "Released with the first official Llama Stack distributions"
      ],
      "source_title": "Llama 3.2: Revolutionizing edge AI and vision with open, customizable models",
      "source_url": "https://ai.meta.com/blog/llama-3-2-connect-2024-vision-edge-mobile-devices/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/09#llama-3-2-vision"
    },
    {
      "slug": "llama-3-2",
      "date": "2024-09-25",
      "model": "Llama 3.2 1B and 3B",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "1B, 3B",
      "model_license": "Llama 3.2 Community License",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "Llama 3.2 1B and 3B are lightweight text-only models designed for edge and mobile devices, supporting a 128K token context window.",
      "highlights": [
        "Officially supports eight languages",
        "Pretraining data cutoff of December 2023",
        "Released alongside the Llama 3.2 Vision models"
      ],
      "source_title": "Llama 3.2: Revolutionizing edge AI and vision with open, customizable models",
      "source_url": "https://ai.meta.com/blog/llama-3-2-connect-2024-vision-edge-mobile-devices/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/09#llama-3-2"
    },
    {
      "slug": "llama-guard-3-vision",
      "date": "2024-09-25",
      "model": "Llama Guard 3 11B Vision",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "11B",
      "model_license": "Llama 3.2 Community License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Llama Guard 3 11B Vision is a safety classifier fine-tuned from Llama 3.2 11B that filters text plus image prompts and text responses, supporting Llama 3.2's image understanding.",
      "highlights": [
        "Fine-tuned from the Llama 3.2 11B pretrained model",
        "Predicts safety labels across 13 MLCommons-based hazard categories"
      ],
      "source_title": "Llama 3.2: Revolutionizing edge AI and vision with open, customizable models",
      "source_url": "https://ai.meta.com/blog/llama-3-2-connect-2024-vision-edge-mobile-devices/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/09#llama-guard-3-vision"
    },
    {
      "slug": "llama-guard-3-1b",
      "date": "2024-09-25",
      "model": "Llama Guard 3 1B",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "1B",
      "model_license": "Llama 3.2 Community License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Llama Guard 3 1B is a compact content safety classifier based on Llama 3.2 1B, with a pruned and quantized version for deployment on mobile devices.",
      "highlights": [
        "Pruned and quantized version shrinks from 2,858 MB to 438 MB",
        "Covers 13 hazard categories in eight languages"
      ],
      "source_title": "Llama 3.2: Revolutionizing edge AI and vision with open, customizable models",
      "source_url": "https://ai.meta.com/blog/llama-3-2-connect-2024-vision-edge-mobile-devices/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/09#llama-guard-3-1b"
    },
    {
      "slug": "molmo",
      "date": "2024-09-25",
      "model": "Molmo",
      "lab": "Ai2",
      "lab_slug": "ai2",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "1B (MolmoE), 7B and 72B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Family of open vision-language models trained without distillation from proprietary VLMs, notable for learning to point at what they perceive in images.",
      "highlights": [
        "Weights, captioning data and fine-tuning data released",
        "Released as MolmoE-1B, Molmo-7B-O, Molmo-7B-D and Molmo-72B",
        "Built from pretrained vision encoders and language-only LLMs"
      ],
      "source_title": "Molmo | Ai2",
      "source_url": "https://allenai.org/blog/molmo",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/09#molmo"
    },
    {
      "slug": "gemini-1-5-pro-002-flash-002",
      "date": "2024-09-24",
      "model": "Gemini 1.5 Pro-002 and 1.5 Flash-002",
      "lab": "Google",
      "lab_slug": "google",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms"
      ],
      "summary": "Google released updated production versions of Gemini 1.5 Pro and 1.5 Flash that built on its experimental releases, with improved overall quality, larger gains in math, long context and vision, and lower latency.",
      "highlights": [
        "Model IDs gemini-1.5-pro-002 and gemini-1.5-flash-002",
        "The gemini-1.5-pro-latest and gemini-1.5-flash-latest aliases were updated to point to the -002 models",
        "Improved the models' ability to follow user instructions while balancing safety"
      ],
      "source_title": "Updated production-ready Gemini models, reduced 1.5 Pro pricing, increased rate limits, and more",
      "source_url": "https://developers.googleblog.com/updated-gemini-models-reduced-15-pro-pricing-increased-rate-limits-and-more/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/09#gemini-1-5-pro-002-flash-002"
    },
    {
      "slug": "llama-3-1-nemotron-51b",
      "date": "2024-09-23",
      "model": "Llama-3.1-Nemotron-51B-Instruct",
      "lab": "NVIDIA",
      "lab_slug": "nvidia",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "51B",
      "model_license": "NVIDIA Open Model License",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "51B model derived from Meta's Llama-3.1-70B with neural architecture search and knowledge distillation, built to run efficiently on a single H100 GPU.",
      "highlights": [
        "Uses block distillation and knowledge distillation",
        "Offered as an NVIDIA NIM microservice",
        "Weights on Hugging Face"
      ],
      "source_title": "Advancing the Accuracy-Efficiency Frontier with Llama-3.1-Nemotron-51B | NVIDIA Technical Blog",
      "source_url": "https://developer.nvidia.com/blog/advancing-the-accuracy-efficiency-frontier-with-llama-3-1-nemotron-51b/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/09#llama-3-1-nemotron-51b"
    },
    {
      "slug": "qwen2-5",
      "date": "2024-09-19",
      "model": "Qwen2.5",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 131072,
      "parameters": "0.5B, 1.5B, 3B, 7B, 14B, 32B, 72B",
      "model_license": "Apache-2.0 (3B and 72B use the Qwen Research License and Qwen License)",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "Qwen2.5 is Qwen's family of dense decoder-only language models from 0.5B to 72B parameters, pre-trained on up to 18T tokens. It added 3B, 14B, and 32B sizes and launched alongside Qwen2.5-Coder and Qwen2.5-Math.",
      "highlights": [
        "Full 131,072-token context with 8,192-token generation for the larger models",
        "All open models except 3B and 72B under Apache 2.0",
        "Qwen2-VL-72B weights were also open-sourced the same day"
      ],
      "source_title": "Qwen2.5: A Party of Foundation Models!",
      "source_url": "https://qwenlm.github.io/blog/qwen2.5/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/09#qwen2-5"
    },
    {
      "slug": "qwen2-5-coder",
      "date": "2024-09-19",
      "model": "Qwen2.5-Coder",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "1.5B, 7B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Qwen2.5-Coder is the next generation of Qwen's open code models, renaming CodeQwen to Qwen-Coder. It continues Qwen2.5 training on 5.5T tokens of code, text-code grounding, and synthetic data.",
      "highlights": [
        "Supports up to 128K tokens of context",
        "Covers 92 programming languages",
        "Released in 1.5B and 7B sizes, with 32B announced as coming soon"
      ],
      "source_title": "Qwen2.5-Coder: Code More, Learn More!",
      "source_url": "https://qwenlm.github.io/blog/qwen2.5-coder/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/09#qwen2-5-coder"
    },
    {
      "slug": "qwen2-5-math",
      "date": "2024-09-19",
      "model": "Qwen2.5-Math",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 4000,
      "parameters": "1.5B, 7B, 72B",
      "model_license": "Apache-2.0 (72B uses the Qwen License Agreement)",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Qwen2.5-Math is Qwen's upgraded math model series with base and instruct models at 1.5B, 7B, and 72B plus the Qwen2.5-Math-RM-72B reward model. It solves English and Chinese math problems with chain-of-thought and tool-integrated reasoning.",
      "highlights": [
        "Supports both chain-of-thought and tool-integrated reasoning",
        "Pre-trained on Qwen Math Corpus v2 with a 4K context length",
        "Qwen does not recommend it for non-math tasks"
      ],
      "source_title": "Qwen2.5-Math: The world's leading open-sourced mathematical LLMs",
      "source_url": "https://qwenlm.github.io/blog/qwen2.5-math/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/09#qwen2-5-math"
    },
    {
      "slug": "mistral-small-2409",
      "date": "2024-09-17",
      "model": "Mistral Small 24.09",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32000,
      "parameters": "22B",
      "model_license": "Mistral Research License",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Upgrade of Mistral Small v24.02 released with open weights under the Mistral Research License as part of a September 2024 update to la Plateforme and le Chat.",
      "highlights": [
        "Self-deployment permitted for non-commercial use",
        "API model ID mistral-small-2409",
        "Released alongside Pixtral 12B access in le Chat"
      ],
      "source_title": "AI in abundance | Mistral AI",
      "source_url": "https://mistral.ai/news/september-24-release",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/09#mistral-small-2409"
    },
    {
      "slug": "nvlm-1-0",
      "date": "2024-09-17",
      "model": "NVLM 1.0 (NVLM-D-72B)",
      "lab": "NVIDIA",
      "lab_slug": "nvidia",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "72B",
      "model_license": "CC-BY-NC-4.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Family of frontier-class multimodal language models from NVIDIA's applied deep learning research team, with the 72B NVLM-D model's weights and training code open-sourced.",
      "highlights": [
        "Handles vision-language tasks while reporting text-only results",
        "Training code released in Megatron-Core",
        "Weights on Hugging Face under CC BY-NC 4.0"
      ],
      "source_title": "NVLM: Open Frontier-Class Multimodal LLMs - NVIDIA ADLR",
      "source_url": "https://research.nvidia.com/labs/adlr/NVLM-1/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/09#nvlm-1-0"
    },
    {
      "slug": "pixtral-12b",
      "date": "2024-09-17",
      "model": "Pixtral 12B",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "12B decoder + 400M vision encoder",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "API",
        "Consumer app"
      ],
      "summary": "Mistral AI's first multimodal model, a 12B decoder based on Mistral NeMo paired with a new vision encoder that accepts images of variable size and aspect ratio.",
      "highlights": [
        "Weights first published September 11, 2024, per Mistral's model card",
        "Supports multiple images within a 128K-token context",
        "Made available in le Chat",
        "API model ID pixtral-12b-2409"
      ],
      "source_title": "Pixtral 12B | Mistral AI",
      "source_url": "https://mistral.ai/news/pixtral-12b",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/09#pixtral-12b"
    },
    {
      "slug": "datagemma",
      "date": "2024-09-12",
      "model": "DataGemma",
      "lab": "Google",
      "lab_slug": "google",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "27B",
      "model_license": "Gemma Terms of Use",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "DataGemma is a pair of Gemma 2-based open models that ground statistical answers in Google's Data Commons knowledge graph to reduce hallucinations. One variant uses retrieval-interleaved generation (RIG), the other retrieval-augmented generation (RAG).",
      "highlights": [
        "Data Commons holds over 240 billion data points from sources such as the UN, WHO and CDC",
        "RIG model annotates statistics with natural language queries to Data Commons",
        "Released on Hugging Face for academic and research use"
      ],
      "source_title": "DataGemma: Using real-world data to address AI hallucinations",
      "source_url": "https://blog.google/technology/ai/google-datagemma-ai-llm/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/09#datagemma"
    },
    {
      "slug": "o1-preview",
      "date": "2024-09-12",
      "model": "OpenAI o1-preview and o1-mini",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "reasoning",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "o1-preview and o1-mini are the first models in OpenAI's o1 series, trained with reinforcement learning to reason through problems before responding. o1-mini is a faster version that OpenAI describes as particularly effective at coding.",
      "highlights": [
        "API model IDs o1-preview-2024-09-12 and o1-mini-2024-09-12, first for usage tier 5 developers",
        "Available to ChatGPT Plus and Team at launch; Enterprise and Edu the following week",
        "API launched without function calling, streaming, or system messages",
        "Both models have since been retired from the API"
      ],
      "source_title": "Introducing OpenAI o1",
      "source_url": "https://openai.com/index/introducing-openai-o1-preview/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/09#o1-preview"
    },
    {
      "slug": "xlam",
      "date": "2024-09-06",
      "model": "xLAM",
      "lab": "Salesforce",
      "lab_slug": "salesforce",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "7B, 8x7B and 8x22B",
      "model_license": "CC-BY-NC-4.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Family of large action models optimized for function calling and AI agents, released for research in sizes for mobile use through high-performance deployments.",
      "highlights": [
        "Variants include xLAM-7b-r, xLAM-8x7b-r and xLAM-8x22b-r",
        "Followed the July 17, 2024 release of the xLAM-1b-fc-r and xLAM-7b-fc-r function-calling models",
        "Released for research under CC BY-NC 4.0"
      ],
      "source_title": "xLAM: A Family of Large Action Models for AI Agents - Salesforce",
      "source_url": "https://www.salesforce.com/blog/large-action-model-ai-agent/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/09#xlam"
    },
    {
      "slug": "deepseek-v2-5",
      "date": "2024-09-05",
      "model": "DeepSeek-V2.5",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "DeepSeek Model License",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "DeepSeek-V2.5 merges DeepSeek-V2-Chat-0628 and DeepSeek-Coder-V2-0724 into one model that combines general conversation and coding abilities. It is better aligned with human preferences and improves writing and instruction following.",
      "highlights": [
        "API users can reach it through either deepseek-chat or deepseek-coder",
        "Strengthened resistance to jailbreak attacks while reducing over-refusal",
        "Weights on Hugging Face"
      ],
      "source_title": "DeepSeek-V2.5: A New Open-Source Model Combining General and Coding Capabilities",
      "source_url": "https://api-docs.deepseek.com/news/news0905",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/09#deepseek-v2-5"
    },
    {
      "slug": "yi-coder",
      "date": "2024-09-05",
      "model": "Yi-Coder",
      "lab": "01.AI",
      "lab_slug": "01-ai",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 131072,
      "parameters": "1.5B, 9B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Yi-Coder is 01.AI's open-source code model series in 1.5B and 9B sizes, released as base and chat versions with long-context support for code understanding and generation.",
      "highlights": [
        "Maximum context length of 128K tokens",
        "Supports 52 major programming languages",
        "Base and chat versions for both sizes on Hugging Face, ModelScope, and wisemodel"
      ],
      "source_title": "GitHub - 01-ai/Yi-Coder: Yi-Coder is a series of open-source code language models that delivers state-of-the-art coding performance with fewer than 10 billion parameters.",
      "source_url": "https://github.com/01-ai/Yi-Coder",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/09#yi-coder"
    },
    {
      "slug": "olmoe",
      "date": "2024-09-04",
      "model": "OLMoE",
      "lab": "Ai2",
      "lab_slug": "ai2",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "MoE: 7B total, 1B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "First mixture-of-experts model in the OLMo family, developed with Contextual AI and released with open data, code, logs and intermediate checkpoints.",
      "highlights": [
        "Trained on 5 trillion tokens",
        "244 pretraining checkpoints released, one every 5,000 steps",
        "Instruct version previewed the Tülu 3 post-training pipeline"
      ],
      "source_title": "OLMoE: An open, small, and state-of-the-art mixture-of-experts model | Ai2",
      "source_url": "https://allenai.org/blog/olmoe-an-open-small-and-state-of-the-art-mixture-of-experts-model-c258432d0514",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/09#olmoe"
    },
    {
      "slug": "minimax-video-01",
      "date": "2024-08-31",
      "model": "MiniMax Video-01",
      "lab": "MiniMax",
      "lab_slug": "minimax",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Video-01 is MiniMax's first AI-native video generation model, producing 720p, 25 fps clips of up to six seconds with cinematic camera movement from a text prompt or a reference image plus text.",
      "highlights": [
        "Supports text-to-video and image-to-video modes",
        "Became the model behind the Hailuo AI video product",
        "Available through the MiniMax API"
      ],
      "source_title": "MiniMax officially releases the Video-01 video generation model",
      "source_url": "https://www.minimax.io/news/video-01",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/08#minimax-video-01"
    },
    {
      "slug": "command-r-08-2024",
      "date": "2024-08-30",
      "model": "Command R and Command R+ (08-2024)",
      "lab": "Cohere",
      "lab_slug": "cohere",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "32B and 104B",
      "model_license": "CC-BY-NC-4.0",
      "available_at_launch": [
        "API",
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "Refreshed versions of Command R and Command R+ with better coding, math, reasoning and latency, plus Structured Outputs and selectable safety modes.",
      "highlights": [
        "Safety modes can be set to strict or contextual",
        "Available on Cohere's API, Amazon SageMaker and OCI",
        "Research weights released as c4ai-command-r-08-2024 and c4ai-command-r-plus-08-2024"
      ],
      "source_title": "Updates to the Command R Series",
      "source_url": "https://cohere.com/blog/command-series-0824",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/08#command-r-08-2024"
    },
    {
      "slug": "qwen2-vl",
      "date": "2024-08-29",
      "model": "Qwen2-VL",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "2B, 7B, 72B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "Qwen2-VL is Qwen's vision-language model series based on Qwen2, with improved understanding of images at various resolutions and videos longer than 20 minutes. Qwen open-sourced the 2B and 7B models and released an API for the 72B model.",
      "highlights": [
        "Understands multilingual text inside images, including most European languages, Japanese, Korean, and Arabic",
        "2B and 7B released under Apache 2.0",
        "Qwen2-VL-72B-Instruct weights followed on September 19, 2024"
      ],
      "source_title": "Qwen2-VL: To See the World More Clearly",
      "source_url": "https://qwenlm.github.io/blog/qwen2-vl/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/08#qwen2-vl"
    },
    {
      "slug": "cogvideox-5b",
      "date": "2024-08-27",
      "model": "CogVideoX-5B",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": "5B",
      "model_license": "CogVideoX License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "CogVideoX-5B is a larger open-weights model in Z.ai's CogVideoX text-to-video series, following CogVideoX-2B. Z.ai added an image-to-video variant, CogVideoX-5B-I2V, on September 19, 2024.",
      "highlights": [
        "Larger sibling of CogVideoX-2B",
        "Image-to-video model CogVideoX-5B-I2V released September 19, 2024",
        "Weights under the CogVideoX License"
      ],
      "source_title": "CogVideo & CogVideoX (GitHub)",
      "source_url": "https://github.com/zai-org/CogVideo",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/08#cogvideox-5b"
    },
    {
      "slug": "gemini-1-5-flash-8b",
      "date": "2024-08-27",
      "model": "Gemini 1.5 Flash-8B",
      "lab": "Google",
      "lab_slug": "google",
      "type": "small",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Gemini 1.5 Flash-8B is a smaller and faster variant of Gemini 1.5 Flash, first released as an experimental model in the Gemini API and made production ready in October 2024 for high-volume and long-context tasks.",
      "highlights": [
        "Experimental model gemini-1.5-flash-8b-exp-0827 released Aug 27, 2024, then updated as gemini-1.5-flash-8b-exp-0924",
        "Stable gemini-1.5-flash-8b-001 declared production ready on Oct 3, 2024",
        "Shut down in the Gemini API on Sep 29, 2025"
      ],
      "source_title": "Release notes - Gemini API | Google AI Studio",
      "source_url": "https://ai.google.dev/gemini-api/docs/changelog",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/08#gemini-1-5-flash-8b"
    },
    {
      "slug": "jamba-1-5",
      "date": "2024-08-22",
      "model": "Jamba 1.5 Mini and Jamba 1.5 Large",
      "lab": "AI21 Labs",
      "lab_slug": "ai21",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "MoE: Mini 52B total, 12B active; Large 398B total, 94B active",
      "model_license": "Jamba Open Model License",
      "available_at_launch": [
        "Open weights",
        "API",
        "Cloud platforms"
      ],
      "summary": "Open SSM-Transformer model family with a 256K effective context window, offered for download and through AI21 Studio and major cloud platforms.",
      "highlights": [
        "Mini handles up to 140K tokens on a single GPU",
        "Available on Google Cloud Vertex AI, Microsoft Azure and NVIDIA NIM at launch",
        "Jamba Open Model License permits research and commercial use"
      ],
      "source_title": "The Jamba 1.5 Open Model Family: The Most Powerful and Efficient Long Context Models",
      "source_url": "https://www.ai21.com/blog/announcing-jamba-model-family/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/08#jamba-1-5"
    },
    {
      "slug": "phi-3-5-mini",
      "date": "2024-08-22",
      "model": "Phi-3.5-mini",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "3.8B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Phi-3.5-mini is a 3.8B open Microsoft model that updates Phi-3-mini with further multilingual pre-training and post-training, improving multilingual, multi-turn conversation and reasoning quality while supporting a 128K context.",
      "highlights": [
        "3.8B dense model with a 128K token context length",
        "Lists support for 23 languages",
        "Offered as a serverless endpoint in Azure AI Studio"
      ],
      "source_title": "Discover the New Multi-Lingual, High-Quality Phi-3.5 SLMs",
      "source_url": "https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/discover-the-new-multi-lingual-high-quality-phi-3-5-slms/4225280",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/08#phi-3-5-mini"
    },
    {
      "slug": "phi-3-5-moe",
      "date": "2024-08-22",
      "model": "Phi-3.5-MoE",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "MoE: 16x3.8B experts, 6.6B active",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Phi-3.5-MoE is Microsoft's first mixture-of-experts Phi model, activating two of 16 experts per token for 6.6B active parameters, with multilingual support and a 128K context length.",
      "highlights": [
        "Mixture-of-experts with 16 experts and 6.6B active parameters",
        "128K token context length",
        "Supports more than 20 languages"
      ],
      "source_title": "Discover the New Multi-Lingual, High-Quality Phi-3.5 SLMs",
      "source_url": "https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/discover-the-new-multi-lingual-high-quality-phi-3-5-slms/4225280",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/08#phi-3-5-moe"
    },
    {
      "slug": "phi-3-5-vision",
      "date": "2024-08-22",
      "model": "Phi-3.5-vision",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "4.2B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Phi-3.5-vision is a 4.2B open multimodal Microsoft model that adds multi-frame image understanding to the Phi-3 vision line, enabling image comparison, multi-image summarization and video summarization.",
      "highlights": [
        "4.2B parameters with a 128K token context length",
        "Adds multi-frame image understanding and reasoning",
        "Not optimized for multilingual use cases"
      ],
      "source_title": "Discover the New Multi-Lingual, High-Quality Phi-3.5 SLMs",
      "source_url": "https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/discover-the-new-multi-lingual-high-quality-phi-3-5-slms/4225280",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/08#phi-3-5-vision"
    },
    {
      "slug": "ideogram-2-0",
      "date": "2024-08-21",
      "model": "Ideogram 2.0",
      "lab": "Ideogram",
      "lab_slug": "ideogram",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Ideogram text-to-image model released free to all users, launched together with the Ideogram iOS app, a beta Ideogram API and Ideogram Search.",
      "highlights": [
        "Five styles: General, Realistic, Design, 3D and Anime",
        "Color palette control, including custom palettes",
        "Design style renders longer, more accurate text"
      ],
      "source_title": "Ideogram on X: Introducing Ideogram 2.0",
      "source_url": "https://x.com/ideogram_ai/status/1826277550798278804",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/08#ideogram-2-0"
    },
    {
      "slug": "mistral-nemo-minitron-8b",
      "date": "2024-08-21",
      "model": "Mistral-NeMo-Minitron 8B",
      "lab": "NVIDIA",
      "lab_slug": "nvidia",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "8B",
      "model_license": "NVIDIA Open Model License",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Small language model created by pruning and distilling the 12B Mistral NeMo model down to 8B parameters while recovering accuracy through retraining.",
      "highlights": [
        "Derived from Mistral NeMo 12B",
        "Offered as an NVIDIA NIM microservice and on Hugging Face",
        "Trained between July 24 and August 10, 2024"
      ],
      "source_title": "NVIDIA Releases Small Language Model With State-of-the-Art Accuracy | NVIDIA Blog",
      "source_url": "https://blogs.nvidia.com/blog/mistral-nemo-minitron-8b-small-language-model/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/08#mistral-nemo-minitron-8b"
    },
    {
      "slug": "xgen-mm-v1-5",
      "date": "2024-08-19",
      "model": "xGen-MM v1.5 (BLIP-3)",
      "lab": "Salesforce",
      "lab_slug": "salesforce",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Series of open multimodal models from Salesforce's xGen-MM family, also known as BLIP-3, including a model that handles interleaved images and text.",
      "highlights": [
        "Built on a Phi-3-mini language backbone",
        "Follows the first xGen-MM (v1) release",
        "Interleave instruct model released under Apache 2.0"
      ],
      "source_title": "Salesforce/xgen-mm-phi3-mini-instruct-interleave-r-v1.5 · Hugging Face",
      "source_url": "https://huggingface.co/Salesforce/xgen-mm-phi3-mini-instruct-interleave-r-v1.5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/08#xgen-mm-v1-5"
    },
    {
      "slug": "deepseek-prover-v1-5",
      "date": "2024-08-16",
      "model": "DeepSeek-Prover-V1.5",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "7B",
      "model_license": "DeepSeek Model License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "DeepSeek-Prover-V1.5 is a 7B open-weights model for theorem proving in Lean 4, pre-trained on DeepSeekMath-Base and refined with reinforcement learning from proof assistant feedback. It improves on DeepSeek-Prover-V1, whose weights were also published.",
      "highlights": [
        "Released as Base, SFT, and RL models",
        "Uses reinforcement learning from proof assistant feedback (RLPAF)",
        "Model use subject to the DeepSeek model license, which supports commercial use"
      ],
      "source_title": "DeepSeek-Prover-V1.5 (GitHub)",
      "source_url": "https://github.com/deepseek-ai/DeepSeek-Prover-V1.5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/08#deepseek-prover-v1-5"
    },
    {
      "slug": "hermes-3",
      "date": "2024-08-15",
      "model": "Hermes 3",
      "lab": "Nous Research",
      "lab_slug": "nous-research",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "8B, 70B and 405B",
      "model_license": "Llama 3 Community License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Steerable instruction-tuned models created by fine-tuning Llama 3.1 8B, 70B and 405B on primarily synthetic data, with long-term context retention and multi-turn chat.",
      "highlights": [
        "Released with the Hermes 3 technical report",
        "Supports complex roleplaying and internal monologue",
        "A Llama 3.2 3B version followed in December 2024"
      ],
      "source_title": "Freedom at the Frontier: Hermes 3 | Nous Research",
      "source_url": "https://nousresearch.com/freedom-at-the-frontier-hermes-3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/08#hermes-3"
    },
    {
      "slug": "runway-gen-3-alpha-turbo",
      "date": "2024-08-14",
      "model": "Gen-3 Alpha Turbo",
      "lab": "Runway",
      "lab_slug": "runway",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Faster, lower-cost version of Gen-3 Alpha that generates seven times faster while matching the original's performance across many use cases.",
      "highlights": [
        "Seven times faster than Gen-3 Alpha",
        "Gained keyframes, extensions, video to video and camera control in later updates",
        "Integrated into apps through the Runway API"
      ],
      "source_title": "Product Updates & Changelog | Runway AI",
      "source_url": "https://runway.com/changelog",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/08#runway-gen-3-alpha-turbo"
    },
    {
      "slug": "grok-2",
      "date": "2024-08-13",
      "model": "Grok-2 and Grok-2 mini",
      "lab": "xAI",
      "lab_slug": "xai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Grok-2 is xAI's successor to Grok-1.5 with improved chat, coding, and reasoning, released in beta for X Premium and Premium+ users alongside Grok-2 mini, a smaller sibling balancing speed and answer quality.",
      "highlights": [
        "An early version was tested on LMSYS Chatbot Arena as sus-column-r",
        "Grok-2 supports both text and vision understanding",
        "xAI experimented with Black Forest Labs' FLUX.1 for image generation on X"
      ],
      "source_title": "Grok-2 Beta Release | SpaceXSI",
      "source_url": "https://x.ai/news/grok-2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/08#grok-2"
    },
    {
      "slug": "qwen2-audio",
      "date": "2024-08-09",
      "model": "Qwen2-Audio",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "7B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Qwen2-Audio is the next version of Qwen-Audio, accepting audio and text input for voice chat and audio analysis. It understands more than 8 languages and dialects.",
      "highlights": [
        "Released as Qwen2-Audio-7B and Qwen2-Audio-7B-Instruct",
        "Supports voice chat and audio analysis modes",
        "Weights on Hugging Face and ModelScope under the Apache 2.0 license"
      ],
      "source_title": "Qwen2-Audio: Chat with Your Voice!",
      "source_url": "https://qwenlm.github.io/blog/qwen2-audio/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/08#qwen2-audio"
    },
    {
      "slug": "qwen2-math",
      "date": "2024-08-08",
      "model": "Qwen2-Math",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "1.5B, 7B, 72B",
      "model_license": "Apache-2.0 (72B uses the Tongyi Qianwen License Agreement)",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Qwen2-Math is Qwen's first series of math-specific language models, initialized from Qwen2 1.5B, 7B, and 72B and pre-trained on a mathematics-specific corpus. It mainly supports English at launch.",
      "highlights": [
        "Released as base and instruct models",
        "Bilingual English and Chinese math models were announced as coming",
        "Smaller sizes under Apache 2.0"
      ],
      "source_title": "Introducing Qwen2-Math",
      "source_url": "https://qwenlm.github.io/blog/qwen2-math/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/08#qwen2-math"
    },
    {
      "slug": "amazon-titan-image-generator-v2",
      "date": "2024-08-06",
      "model": "Amazon Titan Image Generator v2",
      "lab": "Amazon",
      "lab_slug": "amazon",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Amazon Titan Image Generator v2 updates Amazon's image model with image conditioning from reference images, color palette guidance, background removal and subject consistency through fine-tuning.",
      "highlights": [
        "Adds image conditioning with Canny edge or segmentation control",
        "Adds color palette guidance and background removal",
        "Generally available in Amazon Bedrock in US East (N. Virginia) and US West (Oregon)"
      ],
      "source_title": "Amazon Titan Image Generator v2 is now available in Amazon Bedrock",
      "source_url": "https://aws.amazon.com/blogs/aws/amazon-titan-image-generator-v2-is-now-available-in-amazon-bedrock/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/08#amazon-titan-image-generator-v2"
    },
    {
      "slug": "cogvideox-2b",
      "date": "2024-08-06",
      "model": "CogVideoX-2B",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": "2B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "CogVideoX-2B is the first open-weights model in Z.ai's CogVideoX text-to-video series, released with the 3D Causal VAE it uses to reconstruct video. It is a smaller entry-level video generation model.",
      "highlights": [
        "3D Causal VAE open-sourced alongside the model",
        "Weights on Hugging Face under the Apache 2.0 license",
        "Followed by CogVideoX-5B on August 27, 2024"
      ],
      "source_title": "CogVideo & CogVideoX (GitHub)",
      "source_url": "https://github.com/zai-org/CogVideo",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/08#cogvideox-2b"
    },
    {
      "slug": "gpt-4o-2024-08-06",
      "date": "2024-08-06",
      "model": "gpt-4o-2024-08-06",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "gpt-4o-2024-08-06 is a GPT-4o snapshot trained to follow complex JSON schemas, released with Structured Outputs in the API, which constrains output to developer-supplied JSON Schemas via strict function calling or response_format.",
      "highlights": [
        "response_format json_schema works with gpt-4o-2024-08-06 and gpt-4o-mini-2024-07-18",
        "The gpt-4o alias points to this snapshot per the docs model page",
        "128,000-token context and 16,384 max output tokens per the gpt-4o docs page",
        "Fine-tuning for this snapshot became generally available on Aug 20, 2024"
      ],
      "source_title": "Introducing Structured Outputs in the API",
      "source_url": "https://openai.com/index/introducing-structured-outputs-in-the-api/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/08#gpt-4o-2024-08-06"
    },
    {
      "slug": "flux-1-dev",
      "date": "2024-08-01",
      "model": "FLUX.1 [dev]",
      "lab": "Black Forest Labs",
      "lab_slug": "black-forest-labs",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "12B",
      "model_license": "FLUX.1 [dev] Non-Commercial License",
      "available_at_launch": [
        "Open weights",
        "API",
        "Cloud platforms"
      ],
      "summary": "Open-weight, guidance-distilled FLUX.1 model distilled from FLUX.1 [pro], released for non-commercial use with similar quality and prompt adherence while being more efficient than a standard model of its size.",
      "highlights": [
        "Weights on Hugging Face under a non-commercial license",
        "Commercial use available through BFL licensing",
        "12B parameters, same architecture as the FLUX.1 suite"
      ],
      "source_title": "Announcing Black Forest Labs",
      "source_url": "https://bfl.ai/blog/24-08-01-bfl",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/08#flux-1-dev"
    },
    {
      "slug": "flux-1-pro",
      "date": "2024-08-01",
      "model": "FLUX.1 [pro]",
      "lab": "Black Forest Labs",
      "lab_slug": "black-forest-labs",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "12B",
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms"
      ],
      "summary": "Flagship model of the FLUX.1 text-to-image suite that launched alongside Black Forest Labs itself, built on a 12B-parameter hybrid transformer flow-matching architecture and offered through the BFL API and partners.",
      "highlights": [
        "Launched with the FLUX.1 [dev] and [schnell] variants",
        "Hybrid multimodal and parallel diffusion transformer blocks scaled to 12B parameters",
        "Available via the BFL API, Replicate and fal.ai"
      ],
      "source_title": "Announcing Black Forest Labs",
      "source_url": "https://bfl.ai/blog/24-08-01-bfl",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/08#flux-1-pro"
    },
    {
      "slug": "flux-1-schnell",
      "date": "2024-08-01",
      "model": "FLUX.1 [schnell]",
      "lab": "Black Forest Labs",
      "lab_slug": "black-forest-labs",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "12B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "Fastest FLUX.1 variant, a few-step text-to-image model tailored for local development and personal use and released openly under the Apache 2.0 license.",
      "highlights": [
        "Apache 2.0 license",
        "Inference code on GitHub and in Hugging Face Diffusers",
        "Day-one ComfyUI integration"
      ],
      "source_title": "Announcing Black Forest Labs",
      "source_url": "https://bfl.ai/blog/24-08-01-bfl",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/08#flux-1-schnell"
    },
    {
      "slug": "gemma-2-2b",
      "date": "2024-07-31",
      "model": "Gemma 2 2B",
      "lab": "Google",
      "lab_slug": "google",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "2B",
      "model_license": "Gemma Terms of Use",
      "available_at_launch": [
        "Open weights",
        "API",
        "Cloud platforms"
      ],
      "summary": "Gemma 2 2B is a new 2B-parameter version of Gemma built on the Gemma 2 architecture, designed to run efficiently on hardware from phones and laptops to cloud deployments on Vertex AI and GKE.",
      "highlights": [
        "Weights on Kaggle, Hugging Face and Vertex AI Model Garden",
        "Can be tried in Google AI Studio",
        "Released under Gemma terms for research and commercial use"
      ],
      "source_title": "Smaller, Safer, More Transparent: Advancing Responsible AI with Gemma",
      "source_url": "https://developers.googleblog.com/en/smaller-safer-more-transparent-advancing-responsible-ai-with-gemma/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/07#gemma-2-2b"
    },
    {
      "slug": "shieldgemma",
      "date": "2024-07-31",
      "model": "ShieldGemma",
      "lab": "Google",
      "lab_slug": "google",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "2B, 9B, 27B",
      "model_license": "Gemma Terms of Use",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "ShieldGemma is a set of safety classifiers built on Gemma 2 that filter the inputs and outputs of AI models. It detects hate speech, harassment, sexually explicit content and dangerous content.",
      "highlights": [
        "2B suits online classification; 9B and 27B target offline use",
        "Released alongside Gemma 2 2B and the Gemma Scope interpretability tools"
      ],
      "source_title": "Smaller, Safer, More Transparent: Advancing Responsible AI with Gemma",
      "source_url": "https://developers.googleblog.com/en/smaller-safer-more-transparent-advancing-responsible-ai-with-gemma/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/07#shieldgemma"
    },
    {
      "slug": "midjourney-v6-1",
      "date": "2024-07-30",
      "model": "Midjourney V6.1",
      "lab": "Midjourney",
      "lab_slug": "midjourney",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Point release of V6 with more coherent bodies and objects, better image quality and small details, new 2x upscalers, improved text accuracy and a new personalization model, made default at launch.",
      "highlights": [
        "About 25% faster for standard image jobs",
        "Default model until June 16, 2025",
        "Personalization code versioning"
      ],
      "source_title": "Version 6.1",
      "source_url": "https://updates.midjourney.com/version-6-1/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/07#midjourney-v6-1"
    },
    {
      "slug": "meta-sam-2",
      "date": "2024-07-29",
      "model": "SAM 2",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "image",
        "video"
      ],
      "output_modalities": [
        "image",
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "SAM 2 is Meta's next-generation Segment Anything Model, a unified model for real-time promptable object segmentation in both images and videos.",
      "highlights": [
        "Runs at about 44 frames per second",
        "Released with the SA-V dataset of about 51,000 videos under CC BY 4.0",
        "Treats an image as a single-frame video so one architecture handles both"
      ],
      "source_title": "Introducing SAM 2: The next generation of Meta Segment Anything Model for videos and images",
      "source_url": "https://ai.meta.com/blog/segment-anything-2/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/07#meta-sam-2"
    },
    {
      "slug": "deepseek-coder-v2-0724",
      "date": "2024-07-24",
      "model": "DeepSeek-Coder-V2-0724",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "MoE: 236B total, 21B active",
      "model_license": "DeepSeek Model License",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "DeepSeek-Coder-V2-0724 is an updated DeepSeek-Coder-V2 checkpoint with improved general capabilities through alignment optimization. It upgraded the deepseek-coder API model and was later merged into DeepSeek-V2.5.",
      "highlights": [
        "Upgraded the deepseek-coder API model on July 24, 2024",
        "Weights published on Hugging Face as DeepSeek-Coder-V2-Instruct-0724",
        "One of the two models merged to create DeepSeek-V2.5"
      ],
      "source_title": "Change Log | DeepSeek API Docs",
      "source_url": "https://api-docs.deepseek.com/updates",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/07#deepseek-coder-v2-0724"
    },
    {
      "slug": "mistral-large-2",
      "date": "2024-07-24",
      "model": "Mistral Large 2",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "123B",
      "model_license": "Mistral Research License",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Second-generation flagship with 123B parameters and a 128K-token context window, designed for single-node inference and released with weights under the Mistral Research License.",
      "highlights": [
        "Supports dozens of languages and 80+ coding languages",
        "Commercial self-deployment requires a Mistral Commercial License",
        "API model ID mistral-large-2407"
      ],
      "source_title": "Large Enough | Mistral AI",
      "source_url": "https://mistral.ai/news/mistral-large-2407",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/07#mistral-large-2"
    },
    {
      "slug": "llama-3-1",
      "date": "2024-07-23",
      "model": "Llama 3.1",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "8B, 70B, 405B",
      "model_license": "Llama 3.1 Community License",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "Llama 3.1 introduced the 405B model, which Meta described as the largest openly available foundation model, plus upgraded multilingual 8B and 70B models, all with a 128K context window.",
      "highlights": [
        "Supports eight languages",
        "License changed to allow using Llama outputs to improve other models",
        "405B trained on over 15 trillion tokens",
        "Over 25 partners, including AWS, Azure and Google Cloud, offered services on day one"
      ],
      "source_title": "Introducing Llama 3.1: Our most capable models to date",
      "source_url": "https://ai.meta.com/blog/meta-llama-3-1/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/07#llama-3-1"
    },
    {
      "slug": "llama-guard-3",
      "date": "2024-07-23",
      "model": "Llama Guard 3",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "8B",
      "model_license": "Llama 3.1 Community License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Llama Guard 3 is a multilingual content safety classifier fine-tuned from Llama 3.1 8B, released with Llama 3.1 and optimized for search and code interpreter tool calls.",
      "highlights": [
        "Predicts safety labels across 14 categories, adding Code Interpreter Abuse",
        "Supports eight languages"
      ],
      "source_title": "Introducing Llama 3.1: Our most capable models to date",
      "source_url": "https://ai.meta.com/blog/meta-llama-3-1/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/07#llama-guard-3"
    },
    {
      "slug": "prompt-guard",
      "date": "2024-07-23",
      "model": "Prompt Guard",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 512,
      "parameters": "86M",
      "model_license": "Llama 3.1 Community License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Prompt Guard is a small classifier released with Llama 3.1 that labels inputs as benign, prompt injection or jailbreak, built to filter prompt attacks in LLM applications.",
      "highlights": [
        "Fine-tuned from mDeBERTa-v3-base with 86M backbone parameters",
        "512-token context window; longer inputs are split into segments"
      ],
      "source_title": "Introducing Llama 3.1: Our most capable models to date",
      "source_url": "https://ai.meta.com/blog/meta-llama-3-1/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/07#prompt-guard"
    },
    {
      "slug": "cohere-rerank-3-nimble",
      "date": "2024-07-23",
      "model": "Rerank 3 Nimble",
      "lab": "Cohere",
      "lab_slug": "cohere",
      "type": "reranker",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "score"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Faster member of the Rerank 3 series, about three times faster than Rerank 3 while keeping high accuracy, offered in English and multilingual versions.",
      "highlights": [
        "Multilingual version covers 100+ languages",
        "Can rerank documents tens of thousands of tokens long",
        "Handles emails, JSON and tabular data"
      ],
      "source_title": "Introducing Rerank 3 Nimble: Faster Reranking for Enterprise Search and Retrieval-Augmented Generation (RAG) Systems",
      "source_url": "https://cohere.com/blog/rerank-3-nimble",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/07#cohere-rerank-3-nimble"
    },
    {
      "slug": "eleven-turbo-v2-5",
      "date": "2024-07-19",
      "model": "Eleven Turbo v2.5",
      "lab": "ElevenLabs",
      "lab_slug": "elevenlabs",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Low-latency multilingual model that made Hindi, French, Spanish, Mandarin and 27 other languages three times faster and English 25% faster for conversational AI.",
      "highlights": [
        "First support for Vietnamese, Hungarian and Norwegian",
        "Three times faster in non-English languages",
        "Later outclassed by Flash v2.5"
      ],
      "source_title": "Introducing Turbo v2.5",
      "source_url": "https://elevenlabs.io/blog/introducing-turbo-v25",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/07#eleven-turbo-v2-5"
    },
    {
      "slug": "gpt-4o-mini",
      "date": "2024-07-18",
      "model": "GPT-4o mini",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "small",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "GPT-4o mini is OpenAI's cost-efficient small model, supporting text and vision in the API and replacing GPT-3.5 in ChatGPT for Free, Plus, and Team users. It is the first API model to apply OpenAI's instruction hierarchy method.",
      "highlights": [
        "API model ID gpt-4o-mini-2024-07-18",
        "128K context window, up to 16K output tokens, knowledge up to October 2023",
        "Available in the Assistants, Chat Completions, and Batch APIs; fine-tuning launched Jul 23, 2024",
        "ChatGPT Enterprise access began the following week"
      ],
      "source_title": "GPT-4o mini: advancing cost-efficient intelligence",
      "source_url": "https://openai.com/index/gpt-4o-mini-advancing-cost-efficient-intelligence/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/07#gpt-4o-mini"
    },
    {
      "slug": "granite-code-128k",
      "date": "2024-07-18",
      "model": "Granite Code 128K (3B, 8B)",
      "lab": "IBM",
      "lab_slug": "ibm",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "3B and 8B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Long-context versions of the 3B and 8B Granite Code models that extend the context window to 128K tokens, released as base and instruct checkpoints.",
      "highlights": [
        "Extends Granite Code context to 128K tokens",
        "Base and instruct checkpoints for each size",
        "Released under Apache 2.0"
      ],
      "source_title": "ibm-granite/granite-8b-code-instruct-128k · Hugging Face",
      "source_url": "https://huggingface.co/ibm-granite/granite-8b-code-instruct-128k",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/07#granite-code-128k"
    },
    {
      "slug": "mistral-nemo",
      "date": "2024-07-18",
      "model": "Mistral NeMo",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "12B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "12B model built with NVIDIA that offers a 128K-token context window, multilingual skills and a drop-in replacement path for systems using Mistral 7B.",
      "highlights": [
        "Introduced the Tekken tokenizer, trained on more than 100 languages",
        "Trained with quantization awareness for FP8 inference",
        "Base and instruct checkpoints under Apache 2.0"
      ],
      "source_title": "Mistral NeMo | Mistral AI",
      "source_url": "https://mistral.ai/news/mistral-nemo",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/07#mistral-nemo"
    },
    {
      "slug": "codestral-mamba",
      "date": "2024-07-16",
      "model": "Codestral Mamba",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "7.3B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Mamba-architecture code model designed with help from Albert Gu and Tri Dao, offering linear-time inference and tested on in-context retrieval up to 256K tokens.",
      "highlights": [
        "Instructed model with 7,285,403,648 parameters",
        "Released under Apache 2.0",
        "API model ID codestral-mamba-2407"
      ],
      "source_title": "Codestral Mamba | Mistral AI",
      "source_url": "https://mistral.ai/news/codestral-mamba",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/07#codestral-mamba"
    },
    {
      "slug": "mathstral",
      "date": "2024-07-16",
      "model": "Mathstral 7B",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32000,
      "parameters": "7B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "STEM-specialized model built on Mistral 7B for multi-step mathematical reasoning, produced in collaboration with Project Numina and released with open weights.",
      "highlights": [
        "Instructed model for math and STEM subjects",
        "Weights hosted on Hugging Face",
        "Can be fine-tuned with mistral-finetune"
      ],
      "source_title": "MathΣtral | Mistral AI",
      "source_url": "https://mistral.ai/news/mathstral",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/07#mathstral"
    },
    {
      "slug": "smollm",
      "date": "2024-07-16",
      "model": "SmolLM",
      "lab": "Hugging Face",
      "lab_slug": "hugging-face",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "135M, 360M and 1.7B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Family of small language models trained on SmolLM-Corpus, a curated dataset that combines synthetic textbooks, educational Python code and educational web data.",
      "highlights": [
        "Released in 135M, 360M and 1.7B sizes",
        "Training corpus includes Cosmopedia v2 and Python-Edu",
        "Released under Apache 2.0"
      ],
      "source_title": "SmolLM - blazingly fast and remarkably powerful",
      "source_url": "https://huggingface.co/blog/smollm",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/07#smollm"
    },
    {
      "slug": "granite-20b-functioncalling",
      "date": "2024-07-09",
      "model": "Granite 20B FunctionCalling",
      "lab": "IBM",
      "lab_slug": "ibm",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "20B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Function calling model fine-tuned from Granite 20B Code Instruct with a multi-task training approach to bring tool-calling abilities to the Granite family.",
      "highlights": [
        "Based on granite-20b-code-instruct",
        "Trained with a multi-task function calling approach",
        "Released under Apache 2.0"
      ],
      "source_title": "ibm-granite/granite-20b-functioncalling · Hugging Face",
      "source_url": "https://huggingface.co/ibm-granite/granite-20b-functioncalling",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/07#granite-20b-functioncalling"
    },
    {
      "slug": "kolors",
      "date": "2024-07-06",
      "model": "Kolors",
      "lab": "Kuaishou",
      "lab_slug": "kuaishou",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Kolors Model License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Kolors is Kuaishou's large text-to-image model trained on billions of text-image pairs, bilingual in Chinese and English, open-sourced at the 2024 World Artificial Intelligence Conference.",
      "highlights": [
        "Bilingual Chinese and English prompts with a 256-token text context",
        "ControlNet, inpainting, and IP-Adapter variants released in July 2024",
        "Weights free for academic research; commercial use requires registration"
      ],
      "source_title": "GitHub - Kwai-Kolors/Kolors",
      "source_url": "https://github.com/Kwai-Kolors/Kolors",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/07#kolors"
    },
    {
      "slug": "codegeex4",
      "date": "2024-07-05",
      "model": "CodeGeeX4-ALL-9B",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "9B",
      "model_license": "CodeGeeX4 License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "CodeGeeX4-ALL-9B is the open-weights version of Z.ai's CodeGeeX4 code model series, continually trained on GLM-4-9B. A single model supports code completion and generation, code interpreter, web search, function calling, and repository-level code Q&A.",
      "highlights": [
        "128K context length",
        "Available on Hugging Face, ModelScope, WiseModel, and Ollama",
        "Weights under the CodeGeeX4 License"
      ],
      "source_title": "CodeGeeX4: Open Multilingual Code Generation Model (GitHub)",
      "source_url": "https://github.com/zai-org/CodeGeeX4",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/07#codegeex4"
    },
    {
      "slug": "deepseek-v2-chat-0628",
      "date": "2024-06-28",
      "model": "DeepSeek-V2-Chat-0628",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "DeepSeek Model License",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "DeepSeek-V2-Chat-0628 is an upgraded DeepSeek-V2-Chat whose base model was replaced with the Coder-V2 base, improving code generation, math, and reasoning. It also strengthened role-playing.",
      "highlights": [
        "Upgraded the deepseek-chat API model on June 28, 2024",
        "Weights published on Hugging Face as DeepSeek-V2-Chat-0628",
        "Later merged with DeepSeek-Coder-V2-0724 to form DeepSeek-V2.5"
      ],
      "source_title": "Change Log | DeepSeek API Docs",
      "source_url": "https://api-docs.deepseek.com/updates",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/06#deepseek-v2-chat-0628"
    },
    {
      "slug": "gemma-2",
      "date": "2024-06-27",
      "model": "Gemma 2",
      "lab": "Google",
      "lab_slug": "google",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "9B, 27B",
      "model_license": "Gemma Terms of Use",
      "available_at_launch": [
        "Open weights",
        "API",
        "Cloud platforms"
      ],
      "summary": "Gemma 2 is Google's second generation of open models, released in 9B and 27B sizes with a redesigned architecture for performance and inference efficiency. The 27B model can run full-precision inference on a single NVIDIA H100, A100 80GB or TPU host.",
      "highlights": [
        "Redesigned architecture aimed at performance and inference efficiency",
        "27B available to try in Google AI Studio",
        "Weights on Kaggle and Hugging Face under a commercially friendly Gemma license"
      ],
      "source_title": "Gemma 2 is now available to researchers and developers",
      "source_url": "https://blog.google/technology/developers/google-gemma-2/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/06#gemma-2"
    },
    {
      "slug": "doubao-speech-synthesis",
      "date": "2024-06-25",
      "model": "Doubao Speech Synthesis and Voice Cloning",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Volcano Engine released Doubao's large-model speech synthesis and voice cloning models, which infer emotion and intonation from context to produce natural speech and create custom voices from seconds of recorded audio.",
      "highlights": [
        "Voice cloning builds a custom voice from seconds of audio",
        "Speech synthesis predicts emotion and tone from surrounding text",
        "Offered as APIs on Volcano Engine"
      ],
      "source_title": "模型发布公告 2024 - 火山方舟 - 火山引擎",
      "source_url": "https://docs.volcengine.com/docs/ark/model-release-announcement-2024",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/06#doubao-speech-synthesis"
    },
    {
      "slug": "claude-3-5-sonnet",
      "date": "2024-06-20",
      "model": "Claude 3.5 Sonnet",
      "lab": "Anthropic",
      "lab_slug": "anthropic",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 200000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Cloud platforms"
      ],
      "summary": "Claude 3.5 Sonnet is the first release in Anthropic's Claude 3.5 family, running at twice the speed of Claude 3 Opus with stronger coding and vision. It debuted alongside Artifacts, a Claude.ai feature for working with generated content.",
      "highlights": [
        "API model ID claude-3-5-sonnet-20240620",
        "200K-token context window",
        "Free on Claude.ai and the Claude iOS app; also on the Anthropic API, Amazon Bedrock, and Vertex AI"
      ],
      "source_title": "Claude 3.5 Sonnet",
      "source_url": "https://www.anthropic.com/news/claude-3-5-sonnet",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/06#claude-3-5-sonnet"
    },
    {
      "slug": "meta-chameleon",
      "date": "2024-06-18",
      "model": "Chameleon",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "7B, 34B",
      "model_license": "Chameleon Research License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Chameleon is a mixed-modal early-fusion model family from Meta FAIR; Meta released key components of the 7B and 34B models for research with text-only output.",
      "highlights": [
        "Accepts mixed text and image input",
        "Released under a research-only license",
        "The Chameleon image generation model was not released"
      ],
      "source_title": "Sharing new research, models, and datasets from Meta FAIR",
      "source_url": "https://ai.meta.com/blog/meta-fair-research-new-releases/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/06#meta-chameleon"
    },
    {
      "slug": "runway-gen-3-alpha",
      "date": "2024-06-17",
      "model": "Gen-3 Alpha",
      "lab": "Runway",
      "lab_slug": "runway",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "First of Runway's next-generation models trained on new large-scale multimodal infrastructure, trained jointly on videos and images with major gains in fidelity, consistency and motion over Gen-2.",
      "highlights": [
        "Text to video became available on July 1, 2024",
        "Image to video added July 29, 2024",
        "Powers Motion Brush, Advanced Camera Controls and Director Mode"
      ],
      "source_title": "Runway Gen-3 Alpha: AI Video Generation with Temporal Control",
      "source_url": "https://runway.com/research/introducing-gen-3-alpha",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/06#runway-gen-3-alpha"
    },
    {
      "slug": "deepseek-coder-v2",
      "date": "2024-06-14",
      "model": "DeepSeek-Coder-V2",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "MoE: 236B total, 21B active; Lite 16B total, 2.4B active",
      "model_license": "DeepSeek Model License",
      "available_at_launch": [
        "API",
        "Open weights"
      ],
      "summary": "DeepSeek-Coder-V2 is an open-weights mixture-of-experts code model further pre-trained from an intermediate DeepSeek-V2 checkpoint with an additional 6T tokens. It substantially improved coding and math reasoning while keeping general language performance.",
      "highlights": [
        "Released in 236B and 16B Lite sizes, each with Base and Instruct models",
        "128K context length",
        "Upgraded the deepseek-coder API model on June 14, 2024"
      ],
      "source_title": "Change Log | DeepSeek API Docs",
      "source_url": "https://api-docs.deepseek.com/updates",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/06#deepseek-coder-v2"
    },
    {
      "slug": "nemotron-4-340b",
      "date": "2024-06-14",
      "model": "Nemotron-4 340B",
      "lab": "NVIDIA",
      "lab_slug": "nvidia",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "340B",
      "model_license": "NVIDIA Open Model License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Family of open base, instruct and reward models released as a pipeline for generating synthetic data to train other language models, under NVIDIA's open model license.",
      "highlights": [
        "Released as Base, Instruct and Reward models",
        "Downloadable from NVIDIA NGC and Hugging Face",
        "Instruct model trained between December 2023 and May 2024"
      ],
      "source_title": "NVIDIA Releases Open Synthetic Data Generation Pipeline for Training Large Language Models | NVIDIA Blog",
      "source_url": "https://blogs.nvidia.com/blog/nemotron-4-synthetic-data-generation-llm-training/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/06#nemotron-4-340b"
    },
    {
      "slug": "luma-dream-machine",
      "date": "2024-06-12",
      "model": "Dream Machine",
      "lab": "Luma AI",
      "lab_slug": "luma-ai",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Luma's first video generation model, a multimodal transformer trained directly on videos that makes realistic and fantastical clips from text instructions and images, released free to everyone.",
      "highlights": [
        "Built with AWS SageMaker HyperPod H100 infrastructure",
        "Text-to-video and image-to-video",
        "Dream Machine API launched in September 2024"
      ],
      "source_title": "Luma launches Dream Machine (Luma Blog, archived)",
      "source_url": "https://web.archive.org/web/20240615081149/https://blog.lumalabs.ai/p/dream-machine",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/06#luma-dream-machine"
    },
    {
      "slug": "leonardo-phoenix",
      "date": "2024-06-12",
      "model": "Phoenix (preview)",
      "lab": "Leonardo.Ai",
      "lab_slug": "leonardo-ai",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Leonardo.Ai's first foundational image model, built by its own R&D team and launched in preview for all users with a focus on prompt adherence and accurate text in images.",
      "highlights": [
        "Launched alongside Prompt Enhance for shorter prompts",
        "Available through the Leonardo API from August 30, 2024",
        "Built by Leonardo's own R&D team"
      ],
      "source_title": "Leonardo.Ai on X: We're incredibly excited to announce Phoenix, our own foundational model now in preview for ALL users.",
      "source_url": "https://x.com/LeonardoAi/status/1800922658319823279",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/06#leonardo-phoenix"
    },
    {
      "slug": "stable-diffusion-3-medium",
      "date": "2024-06-12",
      "model": "Stable Diffusion 3 Medium",
      "lab": "Stability AI",
      "lab_slug": "stability-ai",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "2B",
      "model_license": "Stability AI Community License",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "First open-weight release from the Stable Diffusion 3 series, a 2B-parameter text-to-image model, also offered through the Stability API with TensorRT and AMD-optimized versions.",
      "highlights": [
        "2B parameters",
        "TensorRT-optimized version developed with NVIDIA",
        "Available on the Stability API powered by Fireworks AI"
      ],
      "source_title": "Announcing the Open Release of Stable Diffusion 3 Medium, Our Most Sophisticated Image Generation Model to Date",
      "source_url": "https://stability.ai/news-updates/stable-diffusion-3-medium",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/06#stable-diffusion-3-medium"
    },
    {
      "slug": "apple-foundation-models",
      "date": "2024-06-10",
      "model": "Apple Foundation Models",
      "lab": "Apple",
      "lab_slug": "apple",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "About 3B (on-device); server size not disclosed",
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Apple introduced its first foundation models for Apple Intelligence: an on-device language model of about 3B parameters and a larger server model running on Private Cloud Compute.",
      "highlights": [
        "Integrated into iOS 18, iPadOS 18 and macOS Sequoia",
        "On-device model uses low-bit palletization averaging 3.7 bits per weight",
        "Uses adapters; a rank-16 on-device adapter is typically tens of megabytes"
      ],
      "source_title": "Introducing Apple's On-Device and Server Foundation Models",
      "source_url": "https://machinelearning.apple.com/research/introducing-apple-foundation-models",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/06#apple-foundation-models"
    },
    {
      "slug": "qwen2",
      "date": "2024-06-07",
      "model": "Qwen2",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "0.5B, 1.5B, 7B, 57B-A14B, 72B",
      "model_license": "Apache-2.0 (72B uses the Tongyi Qianwen License Agreement)",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Qwen2 is Qwen's second-generation model series of base and instruction-tuned models in five sizes, including the Qwen2-57B-A14B mixture-of-experts model. Base models were pre-trained with 32K context.",
      "highlights": [
        "Qwen2-7B-Instruct and Qwen2-72B-Instruct handle up to 128K tokens with YaRN",
        "Includes Qwen2-0.5B, 1.5B, 7B, 57B-A14B, and 72B",
        "Smaller sizes under Apache 2.0"
      ],
      "source_title": "Hello Qwen2",
      "source_url": "https://qwenlm.github.io/blog/qwen2/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/06#qwen2"
    },
    {
      "slug": "kling-1-0",
      "date": "2024-06-06",
      "model": "Kling",
      "lab": "Kuaishou",
      "lab_slug": "kuaishou",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Kling is Kuaishou's first video generation model, a diffusion transformer with a self-developed 3D VAE that generates videos of up to two minutes at 30fps and up to 1080p from text prompts.",
      "highlights": [
        "Kuaishou's July 25, 2024 release dates the opening of trial applications to June 6, 2024",
        "Diffusion transformer with a self-developed 3D VAE for spatiotemporal compression",
        "Initial beta testing ran inside the KuaiYing app; web access and a global beta followed in July 2024"
      ],
      "source_title": "Kuaishou Unveils Proprietary Video Generation Model 'Kling;' Testing Now Available",
      "source_url": "https://ir.kuaishou.com/news-releases/news-release-details/kuaishou-unveils-proprietary-video-generation-model-kling",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/06#kling-1-0"
    },
    {
      "slug": "glm-4-9b",
      "date": "2024-06-05",
      "model": "GLM-4-9B",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "9B",
      "model_license": "GLM-4 License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "GLM-4-9B is the open-weights series from Z.ai's GLM-4 generation, including base, chat, and long-context chat models. GLM-4-9B-Chat supports web browsing, code execution, custom tool calls, and long-text reasoning.",
      "highlights": [
        "GLM-4-9B-Chat supports up to 128K context",
        "GLM-4-9B-Chat-1M supports a 1M-token context",
        "Adds support for 26 languages, including Japanese, Korean, and German"
      ],
      "source_title": "GLM-4 (GitHub)",
      "source_url": "https://github.com/zai-org/GLM-4",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/06#glm-4-9b"
    },
    {
      "slug": "glm-4v-9b",
      "date": "2024-06-05",
      "model": "GLM-4V-9B",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 8000,
      "parameters": "9B",
      "model_license": "GLM-4 License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "GLM-4V-9B is Z.ai's open-weights vision-language model from the GLM-4-9B series, supporting bilingual Chinese and English multi-turn dialogue on 1120 by 1120 images. It was released the same day as GLM-4-9B.",
      "highlights": [
        "High-resolution 1120 by 1120 image input",
        "8K context length",
        "Weights under the GLM-4 License"
      ],
      "source_title": "GLM-4 (GitHub)",
      "source_url": "https://github.com/zai-org/GLM-4",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/06#glm-4v-9b"
    },
    {
      "slug": "stable-audio-open",
      "date": "2024-06-05",
      "model": "Stable Audio Open",
      "lab": "Stability AI",
      "lab_slug": "stability-ai",
      "type": "music",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": "1.1B",
      "model_license": "Stability AI Community License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Open text-to-audio model for generating up to 47 seconds of samples and sound effects such as drum beats, instrument riffs, ambience and foley for music production and sound design.",
      "highlights": [
        "Up to 47 seconds of audio per generation",
        "Weights on Hugging Face",
        "1.1B parameters, per Stability AI's later Stable Audio Open Small announcement"
      ],
      "source_title": "Introducing Stable Audio Open - An Open Source Model for Audio Samples and Sound Design",
      "source_url": "https://stability.ai/news-updates/introducing-stable-audio-open",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/06#stable-audio-open"
    },
    {
      "slug": "elevenlabs-sound-effects",
      "date": "2024-05-31",
      "model": "Text to Sound Effects",
      "lab": "ElevenLabs",
      "lab_slug": "elevenlabs",
      "type": "music",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "ElevenLabs' first sound model, generating sound effects, short instrumental tracks, soundscapes and character voices from text prompts, available to all users.",
      "highlights": [
        "Generates effects, short instrumentals and soundscapes",
        "Aimed at film, TV, games and social creators",
        "Succeeded by SFX v2 and SFX v3"
      ],
      "source_title": "Sound Effects are here",
      "source_url": "https://elevenlabs.io/blog/sound-effects-are-here",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/05#elevenlabs-sound-effects"
    },
    {
      "slug": "codestral",
      "date": "2024-05-29",
      "model": "Codestral (22B)",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32000,
      "parameters": "22B",
      "model_license": "Mistral AI Non-Production License",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Mistral AI's first code model, a 22B open-weight model trained on more than 80 programming languages with fill-in-the-middle completion through a shared instruct and completion endpoint.",
      "highlights": [
        "Trained on 80+ programming languages",
        "Released under the new Mistral AI Non-Production License",
        "API model ID codestral-2405"
      ],
      "source_title": "Codestral | Mistral AI",
      "source_url": "https://mistral.ai/news/codestral",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/05#codestral"
    },
    {
      "slug": "aya-23",
      "date": "2024-05-23",
      "model": "Aya 23",
      "lab": "Cohere",
      "lab_slug": "cohere",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "8B and 35B",
      "model_license": "CC-BY-NC-4.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Multilingual open-weights research models from Cohere For AI that trade Aya 101's breadth for depth, covering 23 languages in 8B and 35B sizes.",
      "highlights": [
        "Builds on the Aya open science initiative",
        "Released with a technical report",
        "Weights published under a non-commercial license"
      ],
      "source_title": "C4AI Launches Aya 23, 8B and 35B Parameter Open Weights Release",
      "source_url": "https://cohere.com/blog/aya23",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/05#aya-23"
    },
    {
      "slug": "mistral-7b-v0-3",
      "date": "2024-05-22",
      "model": "Mistral 7B v0.3",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32000,
      "parameters": null,
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Update of Mistral 7B that extends the vocabulary to 32,768 tokens, released as base and instruct checkpoints and served as open-mistral-7b.",
      "highlights": [
        "Vocabulary extended to 32,768 tokens",
        "32K-token context window",
        "Released under Apache 2.0"
      ],
      "source_title": "Mistral 7B v0.3 | Mistral Docs",
      "source_url": "https://docs.mistral.ai/models/mistral-7b-0-3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/05#mistral-7b-v0-3"
    },
    {
      "slug": "phi-3-small-and-medium",
      "date": "2024-05-21",
      "model": "Phi-3-small and Phi-3-medium",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "7B, 14B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights",
        "API",
        "Cloud platforms"
      ],
      "summary": "Phi-3-small (7B) and Phi-3-medium (14B) are larger instruction-tuned members of Microsoft's Phi-3 family, first announced with Phi-3-mini and made available on Azure and Hugging Face in May 2024.",
      "highlights": [
        "Phi-3-small offers 8K and 128K context variants",
        "Phi-3-medium offers 4K and 128K context variants",
        "First announced as coming soon on April 23, 2024"
      ],
      "source_title": "New models added to the Phi-3 family, available on Microsoft Azure",
      "source_url": "https://azure.microsoft.com/en-us/blog/new-models-added-to-the-phi-3-family-available-on-microsoft-azure/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/05#phi-3-small-and-medium"
    },
    {
      "slug": "phi-3-vision",
      "date": "2024-05-21",
      "model": "Phi-3-vision",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "4.2B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights",
        "API",
        "Cloud platforms"
      ],
      "summary": "Phi-3-vision is the first multimodal model in Microsoft's Phi-3 family, a 4.2B model that combines language and vision to reason over images, extract text and interpret charts and diagrams.",
      "highlights": [
        "4.2B parameters with a 128K token context length",
        "Weights released May 21, 2024",
        "Optimized variants via ONNX Runtime, DirectML and NVIDIA NIM"
      ],
      "source_title": "New models added to the Phi-3 family, available on Microsoft Azure",
      "source_url": "https://azure.microsoft.com/en-us/blog/new-models-added-to-the-phi-3-family-available-on-microsoft-azure/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/05#phi-3-vision"
    },
    {
      "slug": "cogvlm2",
      "date": "2024-05-20",
      "model": "CogVLM2",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 8000,
      "parameters": "19B",
      "model_license": "CogVLM License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "CogVLM2 is Z.ai's next-generation open vision-language model, based on Llama 3 8B, with English and Chinese chat models. It supports image resolutions up to 1344 by 1344.",
      "highlights": [
        "8K content length",
        "A video understanding version, CogVLM2-Video, followed on July 8, 2024",
        "Weights under the CogVLM License"
      ],
      "source_title": "CogVLM2 & CogVLM2-Video (GitHub)",
      "source_url": "https://github.com/zai-org/CogVLM2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/05#cogvlm2"
    },
    {
      "slug": "deepseek-v2-lite",
      "date": "2024-05-16",
      "model": "DeepSeek-V2-Lite",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32000,
      "parameters": "MoE: 16B total, 2.4B active",
      "model_license": "DeepSeek Model License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "DeepSeek-V2-Lite is a smaller mixture-of-experts model in the DeepSeek-V2 family, released as a base model and an SFT chat model. It uses the same economical training and efficient inference architecture as DeepSeek-V2.",
      "highlights": [
        "16B total parameters with 2.4B activated",
        "32K context length",
        "Released as DeepSeek-V2-Lite and DeepSeek-V2-Lite-Chat"
      ],
      "source_title": "DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model (GitHub)",
      "source_url": "https://github.com/deepseek-ai/DeepSeek-V2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/05#deepseek-v2-lite"
    },
    {
      "slug": "doubao-pro-and-lite",
      "date": "2024-05-15",
      "model": "Doubao (Doubao-pro and Doubao-lite)",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "ByteDance launched the Doubao model family on Volcano Engine's Ark platform, with Doubao-pro and Doubao-lite language models in 4K, 32K, and 128K context versions plus a Doubao text embedding model.",
      "highlights": [
        "Doubao-pro offered in 4K, 32K, and 128K context versions",
        "Doubao-lite offered in 4K, 32K, and 128K context versions",
        "Function-calling and role-play variants of Doubao-pro-4k released the same day",
        "Doubao-embedding supports Chinese and English with 4K context"
      ],
      "source_title": "模型发布公告 2024 - 火山方舟 - 火山引擎",
      "source_url": "https://docs.volcengine.com/docs/ark/model-release-announcement-2024",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/05#doubao-pro-and-lite"
    },
    {
      "slug": "gemini-1-5-flash",
      "date": "2024-05-14",
      "model": "Gemini 1.5 Flash",
      "lab": "Google",
      "lab_slug": "google",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms",
        "Consumer app"
      ],
      "summary": "Gemini 1.5 Flash is a lighter-weight member of the Gemini 1.5 family, designed to be fast and efficient to serve at scale while keeping the breakthrough 1 million token long context window.",
      "highlights": [
        "Introduced at Google I/O 2024 in public preview in Google AI Studio and Vertex AI",
        "Optimized for high-volume, high-frequency tasks; the fastest Gemini model served in the API at launch",
        "Stable gemini-1.5-flash-001 listed as generally available in the Gemini API changelog on May 23, 2024"
      ],
      "source_title": "Google Gemini updates: Flash 1.5, Gemma 2 and Project Astra",
      "source_url": "https://blog.google/innovation-and-ai/products/google-gemini-update-flash-ai-assistant-io-2024/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/05#gemini-1-5-flash"
    },
    {
      "slug": "gemini-1-5-pro-2m-context",
      "date": "2024-05-14",
      "model": "Gemini 1.5 Pro (2M context update)",
      "lab": "Google",
      "lab_slug": "google",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 2000000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms",
        "Consumer app"
      ],
      "summary": "At Google I/O 2024, Google released an improved Gemini 1.5 Pro with better code generation, logical reasoning, planning and multi-turn conversation, and extended its context window to 2 million tokens.",
      "highlights": [
        "The 2 million token window was offered by waitlist to API developers and Google Cloud customers at announcement",
        "General availability of the 2M context window was recorded in the Gemini API changelog on June 27, 2024",
        "Also improved audio and image understanding"
      ],
      "source_title": "Google Gemini updates: Flash 1.5, Gemma 2 and Project Astra",
      "source_url": "https://blog.google/innovation-and-ai/products/google-gemini-update-flash-ai-assistant-io-2024/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/05#gemini-1-5-pro-2m-context"
    },
    {
      "slug": "hunyuan-dit",
      "date": "2024-05-14",
      "model": "Hunyuan-DiT",
      "lab": "Tencent",
      "lab_slug": "tencent",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "1.5B",
      "model_license": "Tencent Hunyuan Community License Agreement",
      "available_at_launch": [
        "Open weights",
        "Consumer app"
      ],
      "summary": "Hunyuan-DiT is an open text-to-image diffusion transformer with fine-grained understanding of English and Chinese prompts, using bilingual CLIP and multilingual T5 text encoders and supporting multi-turn, dialogue-driven image generation.",
      "highlights": [
        "1.5B-parameter diffusion transformer",
        "Version 1.1 followed June 13, 2024 and version 1.2 on July 8, 2024",
        "Text-to-image also offered in the Tencent Hunyuan Bot web app"
      ],
      "source_title": "LICENSE.txt · Tencent-Hunyuan/HunyuanDiT at main",
      "source_url": "https://huggingface.co/Tencent-Hunyuan/HunyuanDiT/blob/main/LICENSE.txt",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/05#hunyuan-dit"
    },
    {
      "slug": "imagen-3",
      "date": "2024-05-14",
      "model": "Imagen 3",
      "lab": "Google",
      "lab_slug": "google",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API",
        "Cloud platforms"
      ],
      "summary": "Imagen 3 is Google DeepMind's text-to-image model announced at I/O 2024 as its highest-quality image model at the time. It started in private preview in ImageFX and later reached Vertex AI and the Gemini API.",
      "highlights": [
        "Generally available on Vertex AI for approved users from July 31, 2024 and for all users from December 10, 2024",
        "Reached the Gemini API as imagen-3.0-generate-002 on February 6, 2025",
        "Initially offered in private preview to select creators in ImageFX"
      ],
      "source_title": "New generative media models and tools, built with and for creators",
      "source_url": "https://blog.google/technology/ai/google-generative-ai-veo-imagen-3/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/05#imagen-3"
    },
    {
      "slug": "paligemma",
      "date": "2024-05-14",
      "model": "PaliGemma",
      "lab": "Google",
      "lab_slug": "google",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "3B",
      "model_license": "Gemma Terms of Use",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "PaliGemma is Google's first open vision-language model in the Gemma family, combining the SigLIP vision encoder with a Gemma 2B language model. It targets captioning, visual question answering, OCR, object detection and segmentation.",
      "highlights": [
        "Inspired by PaLI-3; image encoder initialized from SigLIP-So400m",
        "Pretrained, fine-tuned and mixed-task checkpoints at multiple resolutions",
        "Available on GitHub, Hugging Face, Kaggle and Vertex AI Model Garden"
      ],
      "source_title": "Introducing PaliGemma, Gemma 2, and an Upgraded Responsible AI Toolkit",
      "source_url": "https://developers.googleblog.com/en/gemma-family-and-toolkit-expansion-io-2024/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/05#paligemma"
    },
    {
      "slug": "gpt-4o",
      "date": "2024-05-13",
      "model": "GPT-4o",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": 128000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "GPT-4o is OpenAI's omni model, trained end to end across text, vision, and audio and accepting any mix of text, audio, image, and video input. It launched in ChatGPT, including the free tier, and in the API as a text and vision model.",
      "highlights": [
        "API model ID gpt-4o-2024-05-13 at launch",
        "API audio and video capabilities went first to a small group of trusted partners",
        "A new Voice Mode built on GPT-4o was planned as an alpha for ChatGPT Plus",
        "128,000-token context window per the gpt-4o docs model page"
      ],
      "source_title": "Hello GPT-4o",
      "source_url": "https://openai.com/index/hello-gpt-4o/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/05#gpt-4o"
    },
    {
      "slug": "yi-1-5",
      "date": "2024-05-13",
      "model": "Yi-1.5",
      "lab": "01.AI",
      "lab_slug": "01-ai",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "6B, 9B, 34B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Yi-1.5 is an upgraded version of 01.AI's Yi models, continuously pre-trained on a 500B-token high-quality corpus and fine-tuned on 3M samples to improve coding, math, reasoning, and instruction following.",
      "highlights": [
        "Base and chat models in 34B, 9B, and 6B sizes",
        "Context variants of 4K, 16K, and 32K tokens",
        "Code and weights distributed under Apache 2.0"
      ],
      "source_title": "GitHub - 01-ai/Yi-1.5: Yi-1.5 is an upgraded version of Yi, delivering stronger performance in coding, math, reasoning, and instruction-following capability.",
      "source_url": "https://github.com/01-ai/Yi-1.5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/05#yi-1-5"
    },
    {
      "slug": "qwen-max-0428",
      "date": "2024-05-11",
      "model": "Qwen-Max-0428",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Qwen-Max-0428 is an instruction-tuned Qwen model for chat services, larger than the open Qwen1.5-110B. It reached the top 10 of the Chatbot Arena leaderboard shortly after release.",
      "highlights": [
        "Available through the DashScope API",
        "DashScope API became compatible with the OpenAI API format",
        "Outperforms Qwen1.5-110B-Chat on MT-Bench, per Qwen"
      ],
      "source_title": "Notes on Qwen-Max-0428",
      "source_url": "https://qwenlm.github.io/blog/qwen-max-0428/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/05#qwen-max-0428"
    },
    {
      "slug": "amazon-titan-text-premier",
      "date": "2024-05-07",
      "model": "Amazon Titan Text Premier",
      "lab": "Amazon",
      "lab_slug": "amazon",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Amazon Titan Text Premier is the most advanced model in the Titan Text family, optimized for retrieval-augmented generation and agent-based applications through integration with Knowledge Bases and Agents for Amazon Bedrock.",
      "highlights": [
        "32K token context window",
        "Best suited for English-language tasks",
        "Integrates with Knowledge Bases, Agents and Guardrails for Amazon Bedrock"
      ],
      "source_title": "Build RAG and agent-based generative AI applications with new Amazon Titan Text Premier model, available in Amazon Bedrock",
      "source_url": "https://aws.amazon.com/blogs/aws/build-rag-and-agent-based-generative-ai-applications-with-new-amazon-titan-text-premier-model-available-in-amazon-bedrock/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/05#amazon-titan-text-premier"
    },
    {
      "slug": "deepseek-v2",
      "date": "2024-05-06",
      "model": "DeepSeek-V2",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "MoE: 236B total, 21B active",
      "model_license": "DeepSeek Model License",
      "available_at_launch": [
        "API",
        "Consumer app",
        "Open weights"
      ],
      "summary": "DeepSeek-V2 is a mixture-of-experts language model designed for economical training and efficient inference, pre-trained on 8.1T tokens. Compared with DeepSeek 67B, it saves 42.5% of training costs and reduces the KV cache by 93.3%.",
      "highlights": [
        "236B total parameters with 21B activated per token",
        "128K context length",
        "Released as a base model and an RL-tuned chat model"
      ],
      "source_title": "DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model (GitHub)",
      "source_url": "https://github.com/deepseek-ai/DeepSeek-V2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/05#deepseek-v2"
    },
    {
      "slug": "granite-code",
      "date": "2024-05-06",
      "model": "Granite Code (3B, 8B, 20B, 34B)",
      "lab": "IBM",
      "lab_slug": "ibm",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "3B, 8B, 20B and 34B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Family of open code models released as base and instruct versions in four sizes, trained on permissively licensed code and instruction data under Apache 2.0.",
      "highlights": [
        "Context lengths range from 2K tokens for 3B to 8K tokens for 20B and 34B",
        "Instruct versions tuned for logical reasoning and problem solving",
        "Released under Apache 2.0"
      ],
      "source_title": "ibm-granite/granite-34b-code-instruct-8k · Hugging Face",
      "source_url": "https://huggingface.co/ibm-granite/granite-34b-code-instruct-8k",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/05#granite-code"
    },
    {
      "slug": "jamba-instruct",
      "date": "2024-05-02",
      "model": "Jamba-Instruct",
      "lab": "AI21 Labs",
      "lab_slug": "ai21",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Instruction-tuned version of the Jamba open model launched in public preview on the AI21 platform for commercial use with a 256K-token context window.",
      "highlights": [
        "Built on the Jamba SSM-Transformer base model",
        "Later offered on Azure, Amazon Bedrock and Snowflake Cortex",
        "Public preview on the AI21 Platform at launch"
      ],
      "source_title": "Built for the Enterprise: Introducing AI21's Jamba-Instruct Model",
      "source_url": "https://www.ai21.com/blog/announcing-jamba-instruct/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/05#jamba-instruct"
    },
    {
      "slug": "amazon-titan-text-embeddings-v2",
      "date": "2024-04-30",
      "model": "Amazon Titan Text Embeddings V2",
      "lab": "Amazon",
      "lab_slug": "amazon",
      "type": "embedding",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "embedding"
      ],
      "context_window": 8192,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Amazon Titan Text Embeddings V2 is Amazon's second-generation text embedding model, optimized for retrieval-augmented generation, with selectable output dimensions, default normalization and pre-training on more than 100 languages plus code.",
      "highlights": [
        "Accepts up to 8,192 input tokens",
        "Output dimensions of 256, 512 or 1,024",
        "Launched in US East (N. Virginia) and US West (Oregon)"
      ],
      "source_title": "Amazon Titan Text Embeddings V2 now available in Amazon Bedrock, optimized for improving RAG",
      "source_url": "https://aws.amazon.com/blogs/aws/amazon-titan-text-v2-now-available-in-amazon-bedrock-optimized-for-improving-rag/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/04#amazon-titan-text-embeddings-v2"
    },
    {
      "slug": "qwen1-5-110b",
      "date": "2024-04-25",
      "model": "Qwen1.5-110B",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32000,
      "parameters": "110B",
      "model_license": "Tongyi Qianwen License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Qwen1.5-110B is the first 100B-plus model in the Qwen1.5 series, built with the same Transformer decoder architecture as other Qwen1.5 models.",
      "highlights": [
        "32K-token context length",
        "Multilingual, including English, Chinese, French, Spanish, German, Russian, Korean, and Japanese",
        "Weights under the Tongyi Qianwen License Agreement"
      ],
      "source_title": "Qwen1.5-110B: The First 100B+ Model of the Qwen1.5 Series",
      "source_url": "https://qwenlm.github.io/blog/qwen1.5-110b/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/04#qwen1-5-110b"
    },
    {
      "slug": "snowflake-arctic",
      "date": "2024-04-24",
      "model": "Snowflake Arctic",
      "lab": "Snowflake",
      "lab_slug": "snowflake",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "Dense-MoE hybrid: 480B total, 17B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Enterprise-focused open language model with a dense-MoE hybrid architecture that combines a 10B dense transformer with a residual 128-expert MoE layer.",
      "highlights": [
        "Uses top-2 gating across 128 fine-grained experts",
        "Base and instruct checkpoints released under Apache 2.0",
        "Data recipes and research insights open-sourced"
      ],
      "source_title": "Snowflake Arctic - LLM for Enterprise AI",
      "source_url": "https://www.snowflake.com/en/blog/arctic-open-efficient-foundation-language-models-snowflake/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/04#snowflake-arctic"
    },
    {
      "slug": "openelm",
      "date": "2024-04-23",
      "model": "OpenELM",
      "lab": "Apple",
      "lab_slug": "apple",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "270M, 450M, 1.1B, 3B",
      "model_license": "Apple Machine Learning Research Model License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "OpenELM is Apple's family of small open language models that use layer-wise scaling to allocate parameters across transformer layers, released with pretrained and instruction-tuned versions.",
      "highlights": [
        "Pretrained on about 1.8 trillion tokens of public data",
        "Release includes training and evaluation framework, logs and checkpoints",
        "Includes code to convert models to MLX for Apple devices"
      ],
      "source_title": "apple/OpenELM model card",
      "source_url": "https://huggingface.co/apple/OpenELM",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/04#openelm"
    },
    {
      "slug": "phi-3-mini",
      "date": "2024-04-23",
      "model": "Phi-3-mini",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "3.8B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights",
        "API",
        "Cloud platforms"
      ],
      "summary": "Phi-3-mini is a 3.8B instruction-tuned Microsoft small language model and the first of the Phi-3 family, offered in 4K and 128K context variants for on-device, latency-sensitive and cost-constrained uses.",
      "highlights": [
        "Available in 4K and 128K token context variants",
        "Released on Azure AI Studio, Hugging Face and Ollama, and as an NVIDIA NIM microservice",
        "Safety post-training included RLHF and red-teaming"
      ],
      "source_title": "Introducing Phi-3: Redefining what's possible with SLMs",
      "source_url": "https://azure.microsoft.com/en-us/blog/introducing-phi-3-redefining-whats-possible-with-slms/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/04#phi-3-mini"
    },
    {
      "slug": "llama-3",
      "date": "2024-04-18",
      "model": "Llama 3",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 8192,
      "parameters": "8B, 70B",
      "model_license": "Meta Llama 3 Community License",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "Llama 3 is Meta's third-generation open model family, released as 8B and 70B pretrained and instruction-tuned models trained on over 15 trillion tokens.",
      "highlights": [
        "Trained on sequences of 8,192 tokens",
        "Released with Llama Guard 2, Code Shield and CyberSec Eval 2",
        "Meta said models over 400B parameters were still in training"
      ],
      "source_title": "Introducing Meta Llama 3: The most capable openly available LLM to date",
      "source_url": "https://ai.meta.com/blog/meta-llama-3/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/04#llama-3"
    },
    {
      "slug": "llama-guard-2",
      "date": "2024-04-18",
      "model": "Llama Guard 2",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "8B",
      "model_license": "Meta Llama 3 Community License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Llama Guard 2 is an 8B Llama 3 based input and output safeguard model released with Llama 3, using the MLCommons hazard taxonomy.",
      "highlights": [
        "Predicts safety labels on 11 of the 13 MLCommons hazard categories",
        "Can be fine-tuned for different applications"
      ],
      "source_title": "Introducing Meta Llama 3: The most capable openly available LLM to date",
      "source_url": "https://ai.meta.com/blog/meta-llama-3/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/04#llama-guard-2"
    },
    {
      "slug": "mixtral-8x22b",
      "date": "2024-04-17",
      "model": "Mixtral 8x22B",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 64000,
      "parameters": "MoE: 141B total, 39B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Larger sparse mixture-of-experts model released under Apache 2.0, with native function calling, strong math and coding skills and a 64K-token context window.",
      "highlights": [
        "Fluent in English, French, Italian, German and Spanish",
        "Base and instruct checkpoints released",
        "API model ID open-mixtral-8x22b"
      ],
      "source_title": "Cheaper, Better, Faster, Stronger | Mistral AI",
      "source_url": "https://mistral.ai/news/mixtral-8x22b",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/04#mixtral-8x22b"
    },
    {
      "slug": "olmo-1-7",
      "date": "2024-04-17",
      "model": "OLMo 1.7 7B",
      "lab": "Ai2",
      "lab_slug": "ai2",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 4096,
      "parameters": "7B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Updated 7B OLMo model with a longer context window, a new two-stage training procedure and improved data, released with the Dolma 1.7 dataset.",
      "highlights": [
        "Context length raised from 2,048 to 4,096 tokens",
        "Trained on the new Dolma 1.7 data mix",
        "Released under Apache 2.0"
      ],
      "source_title": "OLMo 1.7–7B: A 24 point improvement on MMLU | Ai2",
      "source_url": "https://allenai.org/blog/olmo-1-7-7b-a-24-point-improvement-on-mmlu-92b43f7d269d",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/04#olmo-1-7"
    },
    {
      "slug": "snowflake-arctic-embed",
      "date": "2024-04-16",
      "model": "Arctic Embed",
      "lab": "Snowflake",
      "lab_slug": "snowflake",
      "type": "embedding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "embedding"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Snowflake's family of open text embedding models optimized for retrieval quality at each size, released on Hugging Face under Apache 2.0.",
      "highlights": [
        "Released in xs, s, m, m-long and l sizes",
        "A technical report followed in May 2024",
        "Arctic Embed M v1.5 added compressible vectors in July 2024"
      ],
      "source_title": "Snowflake/snowflake-arctic-embed-m · Hugging Face",
      "source_url": "https://huggingface.co/Snowflake/snowflake-arctic-embed-m",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/04#snowflake-arctic-embed"
    },
    {
      "slug": "codeqwen1-5",
      "date": "2024-04-16",
      "model": "CodeQwen1.5",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 64000,
      "parameters": "7B",
      "model_license": "Tongyi Qianwen License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "CodeQwen1.5-7B is a specialized code model built on the Qwen1.5 language model and pre-trained on around 3T tokens of code-related data.",
      "highlights": [
        "Supports 92 programming languages",
        "Processes up to 64K tokens of context",
        "Weights under the Tongyi Qianwen License Agreement"
      ],
      "source_title": "Code with CodeQwen1.5",
      "source_url": "https://qwenlm.github.io/blog/codeqwen1.5/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/04#codeqwen1-5"
    },
    {
      "slug": "idefics2",
      "date": "2024-04-15",
      "model": "Idefics2",
      "lab": "Hugging Face",
      "lab_slug": "hugging-face",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "8B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "General multimodal model that takes arbitrary sequences of text and images, improving on Idefics1 with a smaller size, an open license and stronger OCR.",
      "highlights": [
        "8B parameters under Apache 2.0",
        "Improved optical character recognition",
        "Base and instruct versions released"
      ],
      "source_title": "Introducing Idefics2: A Powerful 8B Vision-Language Model for the community",
      "source_url": "https://huggingface.co/blog/idefics2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/04#idefics2"
    },
    {
      "slug": "reka-core",
      "date": "2024-04-15",
      "model": "Reka Core",
      "lab": "Reka",
      "lab_slug": "reka",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video",
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Enterprise"
      ],
      "summary": "Reka's largest model at launch, a multimodal language model trained from scratch that accepts text, image, video and audio input and was offered via API, on-premises or on-device deployment.",
      "highlights": [
        "128K context window",
        "Pretrained on text from 32 languages",
        "Launch partners included Snowflake (Cortex), Oracle and AI Singapore",
        "Technical report published alongside the launch"
      ],
      "source_title": "Reka Core: Our Frontier Class Multimodal Language Model",
      "source_url": "https://reka.ai/news/reka-core-our-frontier-class-multimodal-language-model",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/04#reka-core"
    },
    {
      "slug": "cohere-rerank-3",
      "date": "2024-04-11",
      "model": "Rerank 3",
      "lab": "Cohere",
      "lab_slug": "cohere",
      "type": "reranker",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "score"
      ],
      "context_window": 4000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Reranking foundation model for enterprise search and RAG that handles longer documents and semi-structured data such as emails, JSON, code and tables.",
      "highlights": [
        "4K-token context length",
        "Covers 100+ languages",
        "Improved latency and lower total cost of ownership"
      ],
      "source_title": "Rerank 3: Efficient Enterprise Search & Retrieval",
      "source_url": "https://cohere.com/blog/rerank-3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/04#cohere-rerank-3"
    },
    {
      "slug": "codegemma",
      "date": "2024-04-09",
      "model": "CodeGemma",
      "lab": "Google",
      "lab_slug": "google",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "2B, 7B",
      "model_license": "Gemma Terms of Use",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "CodeGemma is a family of Gemma-based open code models for code completion, generation and chat. It includes a 7B pretrained model, a 7B instruction-tuned model and a 2B model for fast local code completion.",
      "highlights": [
        "Trained on an additional 500 billion tokens of mostly English web, math and code data",
        "Available on Kaggle, Hugging Face and Vertex AI Model Garden",
        "CodeGemma 1.1 followed on May 3, 2024"
      ],
      "source_title": "Gemma Family Expands with Models Tailored for Developers and Researchers",
      "source_url": "https://developers.googleblog.com/en/gemma-family-expands-with-models-tailored-for-developers-and-researchers/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/04#codegemma"
    },
    {
      "slug": "recurrentgemma",
      "date": "2024-04-09",
      "model": "RecurrentGemma",
      "lab": "Google",
      "lab_slug": "google",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "2B",
      "model_license": "Gemma Terms of Use",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "RecurrentGemma is a research-oriented open Gemma model that replaces the transformer with recurrent layers and local attention (the Griffin architecture), lowering memory use and keeping sampling speed steady on long sequences.",
      "highlights": [
        "Higher throughput at large batch sizes than transformer-based Gemma",
        "Released on Kaggle with fine-tuning notebooks",
        "A 9B variant followed on June 11, 2024"
      ],
      "source_title": "Gemma Family Expands with Models Tailored for Developers and Researchers",
      "source_url": "https://developers.googleblog.com/en/gemma-family-expands-with-models-tailored-for-developers-and-researchers/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/04#recurrentgemma"
    },
    {
      "slug": "text-embedding-004",
      "date": "2024-04-09",
      "model": "text-embedding-004",
      "lab": "Google",
      "lab_slug": "google",
      "type": "embedding",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "embedding"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms"
      ],
      "summary": "text-embedding-004 is a Google text embedding model released in the Gemini API, offered in Vertex AI as text-embedding-preview-0409, with stronger retrieval performance than Google's earlier embedding models.",
      "highlights": [
        "Supports elastic embedding sizes under 768 dimensions",
        "Released the same day Gemini 1.5 Pro entered public preview in the Gemini API",
        "Shut down in the Gemini API on Jan 14, 2026"
      ],
      "source_title": "Gemini 1.5 Pro Now Available in 180+ Countries; with Native Audio Understanding, System Instructions, JSON Mode and more",
      "source_url": "https://developers.googleblog.com/gemini-15-pro-now-available-in-180-countries-with-native-audio-understanding-system-instructions-json-mode-and-more/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/04#text-embedding-004"
    },
    {
      "slug": "stable-lm-2-12b",
      "date": "2024-04-08",
      "model": "Stable LM 2 12B",
      "lab": "Stability AI",
      "lab_slug": "stability-ai",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "12B",
      "model_license": "Stability AI Community License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "12B-parameter base and instruction-tuned language models trained on 2 trillion tokens in seven languages, released alongside an updated Stable LM 2 1.6B with tool use and function calling.",
      "highlights": [
        "Seven languages including English, Spanish, German and French",
        "Commercial use through a Stability AI Membership",
        "Base and chat models on Hugging Face"
      ],
      "source_title": "Introducing Stable LM 2 12B",
      "source_url": "https://stability.ai/news-updates/introducing-stable-lm-2-12b",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/04#stable-lm-2-12b"
    },
    {
      "slug": "command-r-plus",
      "date": "2024-04-04",
      "model": "Command R+",
      "lab": "Cohere",
      "lab_slug": "cohere",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "104B",
      "model_license": "CC-BY-NC-4.0",
      "available_at_launch": [
        "API",
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "Largest model in Cohere's R series at launch, built for advanced RAG with citations, multi-step tool use and multilingual business workloads.",
      "highlights": [
        "Launched first on Microsoft Azure and Cohere's API",
        "Covers 10 key business languages",
        "Research weights published under a non-commercial license"
      ],
      "source_title": "Introducing Command R+: A Scalable LLM Built for Business | Cohere Blog",
      "source_url": "https://cohere.com/blog/command-r-plus-microsoft-azure",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/04#command-r-plus"
    },
    {
      "slug": "stable-audio-2",
      "date": "2024-04-03",
      "model": "Stable Audio 2.0",
      "lab": "Stability AI",
      "lab_slug": "stability-ai",
      "type": "music",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Second-generation Stable Audio model producing full tracks with coherent structure up to three minutes long at 44.1 kHz stereo from a single prompt, and adding audio-to-audio generation.",
      "highlights": [
        "Trained exclusively on licensed data from the AudioSparx library",
        "Free to use on the Stable Audio website at launch",
        "API availability followed later"
      ],
      "source_title": "Introducing Stable Audio 2.0",
      "source_url": "https://stability.ai/news-updates/stable-audio-2-0",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/04#stable-audio-2"
    },
    {
      "slug": "qwen1-5-32b",
      "date": "2024-04-02",
      "model": "Qwen1.5-32B",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32000,
      "parameters": "32B",
      "model_license": "Tongyi Qianwen License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Qwen1.5-32B and Qwen1.5-32B-Chat are mid-size additions to the Qwen1.5 series, aimed at balancing performance, efficiency, and memory footprint at about 30 billion parameters.",
      "highlights": [
        "Released as base and chat models",
        "Uses the same 32K context as other Qwen1.5 models",
        "Weights under the Tongyi Qianwen License Agreement"
      ],
      "source_title": "Qwen1.5-32B: Fitting the Capstone of the Qwen1.5 Language Model Series",
      "source_url": "https://qwenlm.github.io/blog/qwen1.5-32b/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/04#qwen1-5-32b"
    },
    {
      "slug": "grok-1-5",
      "date": "2024-03-28",
      "model": "Grok-1.5",
      "lab": "xAI",
      "lab_slug": "xai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Grok-1.5 is xAI's update to Grok-1 with improved reasoning and problem solving and long-context understanding of up to 128K tokens, 16 times the previous context length.",
      "highlights": [
        "128,000-token context window",
        "Initially for early testers and existing Grok users on X",
        "Company timeline lists Grok-1.5 on X on May 15, 2024"
      ],
      "source_title": "Announcing Grok-1.5 | SpaceXSI",
      "source_url": "https://x.ai/news/grok-1.5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/03#grok-1-5"
    },
    {
      "slug": "jamba",
      "date": "2024-03-28",
      "model": "Jamba",
      "lab": "AI21 Labs",
      "lab_slug": "ai21",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 256000,
      "parameters": "MoE: 52B total, 12B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "The first production-grade model based on Mamba, a hybrid of structured state space and transformer layers with mixture-of-experts, released as an open base model.",
      "highlights": [
        "256K-token context window",
        "Uses 12B of 52B parameters at inference",
        "An instruction-tuned Jamba-Instruct followed on the AI21 platform"
      ],
      "source_title": "Introducing Jamba: AI21's Groundbreaking SSM-Transformer Model",
      "source_url": "https://www.ai21.com/blog/announcing-jamba/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/03#jamba"
    },
    {
      "slug": "qwen1-5-moe",
      "date": "2024-03-28",
      "model": "Qwen1.5-MoE-A2.7B",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "MoE: 2.7B active",
      "model_license": "Tongyi Qianwen License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Qwen1.5-MoE-A2.7B is Qwen's first mixture-of-experts model, with 2.7B activated parameters, created by repurposing Qwen-1.8B. Qwen reports it matches leading 7B models.",
      "highlights": [
        "Contains 2.0B non-embedding parameters, about one-third of Qwen1.5-7B",
        "Upcycled from the dense Qwen-1.8B model",
        "Weights under the Tongyi Qianwen License Agreement"
      ],
      "source_title": "Qwen1.5-MoE: Matching 7B Model Performance with 1/3 Activated Parameters",
      "source_url": "https://qwenlm.github.io/blog/qwen-moe/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/03#qwen1-5-moe"
    },
    {
      "slug": "dbrx",
      "date": "2024-03-27",
      "model": "DBRX",
      "lab": "Databricks",
      "lab_slug": "databricks",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32000,
      "parameters": "MoE: 132B total, 36B active",
      "model_license": "Databricks Open Model License",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Open general-purpose language model from Databricks built on a fine-grained mixture-of-experts architecture, released as DBRX Base and DBRX Instruct for download and API use.",
      "highlights": [
        "Pretrained on 12 trillion tokens of curated data",
        "Inference up to twice as fast as LLaMA2-70B",
        "Offered to Databricks customers through APIs at launch"
      ],
      "source_title": "Introducing DBRX: A New State-of-the-Art Open LLM | Databricks Blog",
      "source_url": "https://www.databricks.com/blog/introducing-dbrx-new-state-art-open-llm",
      "retired_date": "2025-12-19",
      "retirement_note": "Databricks retired DBRX and DBRX Instruct from pay-per-token serving on April 30, 2025 and from provisioned throughput on December 19, 2025.",
      "retirement_source_url": "https://docs.databricks.com/aws/en/machine-learning/retired-models-policy",
      "permalink": "https://theaiengineering.com/releases/2024/03#dbrx"
    },
    {
      "slug": "stable-code-instruct-3b",
      "date": "2024-03-25",
      "model": "Stable Code Instruct 3B",
      "lab": "Stability AI",
      "lab_slug": "stability-ai",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "3B",
      "model_license": "Stability AI Community License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Instruction-tuned version of Stable Code 3B that adds natural language interaction for code generation and software development tasks, focusing on Python, JavaScript, Java, C, C++ and Go.",
      "highlights": [
        "Built on Stable Code 3B",
        "Weights and code on Hugging Face",
        "Commercial use through a Stability AI Membership"
      ],
      "source_title": "Introducing Stable Code Instruct 3B",
      "source_url": "https://stability.ai/news-updates/introducing-stable-code-instruct-3b",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/03#stable-code-instruct-3b"
    },
    {
      "slug": "suno-v3",
      "date": "2024-03-21",
      "model": "Suno v3",
      "lab": "Suno",
      "lab_slug": "suno",
      "type": "music",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Suno's first model capable of radio-quality music, generating full two-minute songs in seconds, released to all users after testing as v3 Alpha with paid subscribers.",
      "highlights": [
        "Full two-minute songs",
        "Tested as v3 Alpha with Pro and Premier members",
        "Available to all users at launch"
      ],
      "source_title": "Introducing v3",
      "source_url": "https://suno.com/blog/v3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/03#suno-v3"
    },
    {
      "slug": "grok-1-open-weights",
      "date": "2024-03-17",
      "model": "Grok-1 (open weights)",
      "lab": "xAI",
      "lab_slug": "xai",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 8192,
      "parameters": "314B (MoE)",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "xAI released the base model weights and network architecture of Grok-1, a 314B-parameter mixture-of-experts model trained from scratch, as the raw pretraining checkpoint without fine-tuning for dialogue.",
      "highlights": [
        "Mixture of 8 experts with 2 used per token (25% of weights active)",
        "Pretraining concluded in October 2023 on a custom JAX and Rust stack",
        "Weights on Hugging Face (xai-org/grok-1) and code on GitHub (xai-org/grok-1)",
        "8,192-token maximum context per the GitHub README"
      ],
      "source_title": "Open Release of Grok-1 | SpaceXSI",
      "source_url": "https://x.ai/news/grok-os",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/03#grok-1-open-weights"
    },
    {
      "slug": "recraft-v2",
      "date": "2024-03-13",
      "model": "Recraft V2",
      "lab": "Recraft",
      "lab_slug": "recraft",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Recraft's image model built for designers, generating both vector and raster images in consistent styles, compared by Recraft against Midjourney v6, DALL-E 3 and open models.",
      "highlights": [
        "Generates vector and raster images",
        "Designed for brand-consistent graphics",
        "Available in the Recraft app"
      ],
      "source_title": "Recraft V2: First AI Model Built for Designers",
      "source_url": "https://www.recraft.ai/blog/recraft-v2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/03#recraft-v2"
    },
    {
      "slug": "command-r",
      "date": "2024-03-11",
      "model": "Command R",
      "lab": "Cohere",
      "lab_slug": "cohere",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": "35B",
      "model_license": "CC-BY-NC-4.0",
      "available_at_launch": [
        "API",
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "Scalable generative model built for retrieval-augmented generation and tool use, with inline citations and a 128K-token context window, plus weights for research use.",
      "highlights": [
        "Strong in 10 business languages",
        "Designed to pair with Cohere Embed and Rerank",
        "Weights on Hugging Face under a non-commercial license"
      ],
      "source_title": "Command R: RAG at Production Scale",
      "source_url": "https://cohere.com/blog/command-r",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/03#command-r"
    },
    {
      "slug": "deepseek-vl",
      "date": "2024-03-11",
      "model": "DeepSeek-VL",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "1.3B, 7B",
      "model_license": "DeepSeek Model License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "DeepSeek-VL is a vision-language model family released in 1.3B and 7B sizes with base and chat variants. The 7B model uses a hybrid SigLIP and SAM vision encoder that supports 1024 by 1024 image input.",
      "highlights": [
        "Four models: 1.3B and 7B, each as base and chat",
        "DeepSeek-VL-7B-base is trained on around 400B vision-language tokens",
        "Supports commercial use under the DeepSeek Model License"
      ],
      "source_title": "DeepSeek-VL: Towards Real-World Vision-Language Understanding (GitHub)",
      "source_url": "https://github.com/deepseek-ai/DeepSeek-VL",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/03#deepseek-vl"
    },
    {
      "slug": "hermes-2-pro-mistral-7b",
      "date": "2024-03-11",
      "model": "Hermes-2-Pro-Mistral-7B",
      "lab": "Nous Research",
      "lab_slug": "nous-research",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "7B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Hermes 2 Pro model from Nous Research built on Mistral 7B, which Nous describes as a master function-calling model, released under Apache 2.0.",
      "highlights": [
        "Focused on function calling",
        "Built on Mistral 7B",
        "Later followed by Llama 3-based Hermes 2 Pro models"
      ],
      "source_title": "Releases | Nous Research",
      "source_url": "https://nousresearch.com/releases",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/03#hermes-2-pro-mistral-7b"
    },
    {
      "slug": "inflection-2-5",
      "date": "2024-03-07",
      "model": "Inflection-2.5",
      "lab": "Inflection AI",
      "lab_slug": "inflection-ai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Inflection-2.5 is Inflection AI's upgraded in-house model for its Pi assistant, rolled out to all Pi users with improved coding and mathematics over Inflection-1 and launched alongside real-time web search in Pi.",
      "highlights": [
        "Available to all Pi users at pi.ai, on iOS, on Android and in a new desktop app",
        "Pi added real-time web search alongside the upgrade",
        "On March 19, 2024, Inflection said it would host Inflection-2.5 on Microsoft Azure for API access"
      ],
      "source_title": "Inflection-2.5: meet the world's best personal AI",
      "source_url": "https://web.archive.org/web/20240531050523/https://inflection.ai/inflection-2-5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/03#inflection-2-5"
    },
    {
      "slug": "yi-9b",
      "date": "2024-03-06",
      "model": "Yi-9B",
      "lab": "01.AI",
      "lab_slug": "01-ai",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 4096,
      "parameters": "9B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Yi-9B is an open bilingual base model from 01.AI, continuously trained from Yi-6B on 0.8T tokens and positioned as the Yi series model strongest at coding and math.",
      "highlights": [
        "Default context window of 4K tokens",
        "Yi-9B-200K long-context variant was open-sourced on March 16, 2024",
        "Weights on Hugging Face, ModelScope, and wisemodel"
      ],
      "source_title": "GitHub - 01-ai/Yi: A series of large language models trained from scratch by developers @01-ai",
      "source_url": "https://github.com/01-ai/Yi",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/03#yi-9b"
    },
    {
      "slug": "claude-3",
      "date": "2024-03-04",
      "model": "Claude 3 Opus, Sonnet, and Haiku",
      "lab": "Anthropic",
      "lab_slug": "anthropic",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 200000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app",
        "Cloud platforms"
      ],
      "summary": "The Claude 3 family comprises Claude 3 Haiku, Sonnet, and Opus in ascending order of capability. It was the first Anthropic generation with vision input, and Anthropic reported fewer unnecessary refusals and improved accuracy over Claude 2.1.",
      "highlights": [
        "200K-token context window at launch; inputs over 1M tokens offered to select customers",
        "Opus and Sonnet available at launch on claude.ai and the Claude API; Haiku followed on March 13, 2024",
        "Sonnet on Amazon Bedrock at launch and in private preview on Google Cloud Vertex AI",
        "Claude API became generally available in 159 countries"
      ],
      "source_title": "Introducing the next generation of Claude",
      "source_url": "https://www.anthropic.com/news/claude-3-family",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/03#claude-3"
    },
    {
      "slug": "ideogram-1-0",
      "date": "2024-02-28",
      "model": "Ideogram 1.0",
      "lab": "Ideogram",
      "lab_slug": "ideogram",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Ideogram text-to-image model focused on text rendering, photorealism and prompt adherence, released to all users with Magic Prompt to help write detailed prompts.",
      "highlights": [
        "Paid plans added private generation, image upload and Ideogram Editor",
        "Magic Prompt expands short prompts into detailed ones"
      ],
      "source_title": "Ideogram on X: Introducing Ideogram 1.0: the most advanced text-to-image model",
      "source_url": "https://x.com/ideogram_ai/status/1762881278955700270",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/02#ideogram-1-0"
    },
    {
      "slug": "mistral-large",
      "date": "2024-02-26",
      "model": "Mistral Large",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms"
      ],
      "summary": "Mistral AI's flagship proprietary model at launch, fluent in five European languages with native function calling, offered on la Plateforme and Microsoft Azure.",
      "highlights": [
        "32K-token context window",
        "Native function calling and JSON output mode",
        "Azure was the first distribution partner",
        "API model ID mistral-large-2402"
      ],
      "source_title": "Au Large | Mistral AI",
      "source_url": "https://mistral.ai/news/mistral-large",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/02#mistral-large"
    },
    {
      "slug": "mistral-small-2402",
      "date": "2024-02-26",
      "model": "Mistral Small (24.02)",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Latency- and cost-optimized proprietary model released alongside Mistral Large, positioned as an intermediary between Mistral's open-weight offering and its flagship, with RAG and function calling support.",
      "highlights": [
        "Supports function calling and JSON mode",
        "API model ID mistral-small-2402",
        "The earlier mistral-small endpoint was renamed open-mixtral-8x7b"
      ],
      "source_title": "Au Large | Mistral AI",
      "source_url": "https://mistral.ai/news/mistral-large",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/02#mistral-small-2402"
    },
    {
      "slug": "stable-diffusion-3",
      "date": "2024-02-22",
      "model": "Stable Diffusion 3",
      "lab": "Stability AI",
      "lab_slug": "stability-ai",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "800M to 8B",
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Stable Diffusion 3 suite combining a diffusion transformer architecture with flow matching, ranging from 800M to 8B parameters, announced in early preview with improved multi-subject prompts and spelling.",
      "highlights": [
        "Announced with a waitlist for an early preview",
        "Stable Diffusion 3 and SD3 Turbo became available in the Stability API on April 17, 2024",
        "Multimodal Diffusion Transformer uses separate weights for image and language"
      ],
      "source_title": "Stable Diffusion 3",
      "source_url": "https://stability.ai/news-updates/stable-diffusion-3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/02#stable-diffusion-3"
    },
    {
      "slug": "gemma",
      "date": "2024-02-21",
      "model": "Gemma",
      "lab": "Google",
      "lab_slug": "google",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 8192,
      "parameters": "2B, 7B",
      "model_license": "Gemma Terms of Use",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "Gemma is Google's first family of lightweight open models, released in 2B and 7B sizes, each with pre-trained and instruction-tuned variants. Its terms of use permit responsible commercial usage and distribution by organizations of any size.",
      "highlights": [
        "Trained with an 8,192-token context length",
        "Weights on Kaggle and Hugging Face; deployable on Vertex AI and GKE",
        "Gemma 1.1 update released April 5, 2024"
      ],
      "source_title": "Gemma: Introducing new state-of-the-art open models",
      "source_url": "https://blog.google/technology/developers/gemma-open-models/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/02#gemma"
    },
    {
      "slug": "gemini-1-5-pro",
      "date": "2024-02-15",
      "model": "Gemini 1.5 Pro",
      "lab": "Google",
      "lab_slug": "google",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 1000000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms",
        "Consumer app"
      ],
      "summary": "Gemini 1.5 Pro was the first model in Google's second Gemini generation, a mid-size multimodal model that performed at a similar level to 1.0 Ultra and introduced an experimental context window of up to 1 million tokens.",
      "highlights": [
        "Launched with a standard 128,000-token window and up to 1 million tokens for a limited group in private preview",
        "Public preview in the Gemini API in 180+ countries began Apr 9, 2024, adding native audio understanding",
        "Stable gemini-1.5-pro-001 listed as generally available in the Gemini API changelog on May 23, 2024"
      ],
      "source_title": "Introducing Gemini 1.5, Google's next-generation AI model",
      "source_url": "https://blog.google/innovation-and-ai/products/google-gemini-next-generation-model-february-2024/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/02#gemini-1-5-pro"
    },
    {
      "slug": "aya-101",
      "date": "2024-02-13",
      "model": "Aya 101",
      "lab": "Cohere",
      "lab_slug": "cohere",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "13B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Open-source massively multilingual research model from Cohere For AI covering 101 languages, released with the largest multilingual instruction fine-tuning dataset at the time.",
      "highlights": [
        "Built with more than 3,000 researchers from 119 countries",
        "Released with a 513 million instance dataset covering 114 languages",
        "Expands coverage to more than 50 previously unserved languages"
      ],
      "source_title": "C4AI Launches Aya, an LLM Covering More Than 100 Languages",
      "source_url": "https://cohere.com/blog/aya",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/02#aya-101"
    },
    {
      "slug": "reka-flash",
      "date": "2024-02-12",
      "model": "Reka Flash and Reka Edge",
      "lab": "Reka",
      "lab_slug": "reka",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "Flash: 21B; Edge: 7B",
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Two multimodal, multilingual models: Reka Flash, a 21B model trained from scratch for fast, high-quality workloads, and the compact 7B Reka Edge for on-device and local use, both in public beta.",
      "highlights": [
        "Pretrained on text from 32 languages",
        "Image and video understanding, including video captioning and video question answering",
        "Instruction tuned, then trained with RLHF (PPO) using Reka Flash as the reward model",
        "Public beta in the Reka Playground at launch"
      ],
      "source_title": "Reka Flash: Efficient and Capable Multimodal Language Models",
      "source_url": "https://reka.ai/news/reka-flash-efficient-and-capable-multimodal-language-models",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/02#reka-flash"
    },
    {
      "slug": "stable-cascade",
      "date": "2024-02-12",
      "model": "Stable Cascade",
      "lab": "Stability AI",
      "lab_slug": "stability-ai",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "Stage C: 1B or 3.6B",
      "model_license": "Stability AI Non-Commercial Research Community License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Research-preview text-to-image model built on the Würstchen architecture, using a three-stage design that decouples text-conditional generation from decoding to make fine-tuning more efficient.",
      "highlights": [
        "Stage C released in 1B and 3.6B parameter versions",
        "Non-commercial license",
        "Training and inference code on GitHub"
      ],
      "source_title": "Introducing Stable Cascade",
      "source_url": "https://stability.ai/news-updates/introducing-stable-cascade",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/02#stable-cascade"
    },
    {
      "slug": "gemini-1-0-ultra",
      "date": "2024-02-08",
      "model": "Gemini 1.0 Ultra",
      "lab": "Google",
      "lab_slug": "google",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms",
        "Consumer app"
      ],
      "summary": "Gemini 1.0 Ultra was the largest and most capable model in Google's first Gemini generation. It reached the public through Gemini Advanced on the day Bard was renamed Gemini.",
      "highlights": [
        "Announced Dec 6, 2023 as the largest of the three Gemini 1.0 sizes (Ultra, Pro and Nano)",
        "Launched in Gemini Advanced on Feb 8, 2024, alongside a new Gemini mobile app",
        "Developers and Cloud customers could begin building with 1.0 Ultra through the Gemini API and Vertex AI from Feb 15, 2024"
      ],
      "source_title": "Google Bard is now Gemini: How to try Ultra 1.0 and new mobile app",
      "source_url": "https://blog.google/products-and-platforms/products/gemini/bard-gemini-advanced-app/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/02#gemini-1-0-ultra"
    },
    {
      "slug": "deepseekmath-7b",
      "date": "2024-02-05",
      "model": "DeepSeekMath 7B",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "reasoning",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "7B",
      "model_license": "DeepSeek Model License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "DeepSeekMath 7B is a math-specialized model initialized from DeepSeek-Coder-v1.5 7B and further pre-trained on 500B tokens of math-related, natural language, and code data. Its RL variant was trained with DeepSeek's Group Relative Policy Optimization (GRPO).",
      "highlights": [
        "Released as Base, Instruct, and RL models",
        "Introduced the GRPO reinforcement learning algorithm",
        "Supports commercial use under the DeepSeek Model License"
      ],
      "source_title": "deepseek-ai/deepseek-math-7b-rl (Hugging Face)",
      "source_url": "https://huggingface.co/deepseek-ai/deepseek-math-7b-rl",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/02#deepseekmath-7b"
    },
    {
      "slug": "qwen1-5",
      "date": "2024-02-04",
      "model": "Qwen1.5",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32768,
      "parameters": "0.5B, 1.8B, 4B, 7B, 14B, 72B",
      "model_license": "Tongyi Qianwen License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Qwen1.5 is the next iteration of the Qwen series, open-sourcing base and chat models across multiple sizes with quantized versions. It improved the developer experience with Hugging Face Transformers integration.",
      "highlights": [
        "All models support a context length of up to 32,768 tokens",
        "Released just before the Chinese New Year with GPTQ, AWQ, and GGUF quantized models; code merged into Hugging Face Transformers",
        "Weights under the Tongyi Qianwen License Agreement"
      ],
      "source_title": "Introducing Qwen1.5",
      "source_url": "https://qwenlm.github.io/blog/qwen1.5/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/02#qwen1-5"
    },
    {
      "slug": "olmo",
      "date": "2024-02-01",
      "model": "OLMo",
      "lab": "Ai2",
      "lab_slug": "ai2",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 2048,
      "parameters": "1B and 7B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Ai2's first truly open language model, released with its full pretraining data, training code, model weights and evaluation tools so researchers can study how it was built.",
      "highlights": [
        "Pretrained on the open Dolma corpus of about three trillion tokens",
        "Built with partners including the Kempner Institute, AMD and the LUMI supercomputer",
        "Available on Hugging Face and GitHub"
      ],
      "source_title": "Hello OLMo: A truly open LLM | Ai2",
      "source_url": "https://allenai.org/blog/hello-olmo-a-truly-open-llm-43f7e7359222",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/02#olmo"
    },
    {
      "slug": "code-llama-70b",
      "date": "2024-01-29",
      "model": "Code Llama 70B",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "70B",
      "model_license": "Llama 2 Community License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Code Llama 70B is the largest model in the Code Llama family, released in base, Python and Instruct versions and free for research and commercial use.",
      "highlights": [
        "Same three versions as the original release: Code Llama, Code Llama - Python and Code Llama - Instruct",
        "Free for research and commercial use"
      ],
      "source_title": "Introducing Code Llama, a state-of-the-art large language model for coding",
      "source_url": "https://ai.meta.com/blog/code-llama-large-language-model-coding/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/01#code-llama-70b"
    },
    {
      "slug": "sfr-embedding-mistral",
      "date": "2024-01-26",
      "model": "SFR-Embedding-Mistral",
      "lab": "Salesforce",
      "lab_slug": "salesforce",
      "type": "embedding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "embedding"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "CC-BY-NC-4.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Text embedding model from Salesforce AI Research built on E5-mistral-7b-instruct and Mistral-7B-v0.1, fine-tuned with multi-task training for retrieval, clustering and related tasks.",
      "highlights": [
        "Ranked first on the MTEB benchmark at release, averaging 67.6 across 56 datasets",
        "Raised retrieval performance from 56.9 for E5-mistral-7b-instruct to 59.0",
        "Released for research use under CC BY-NC 4.0"
      ],
      "source_title": "SFR-Embedding-Mistral: Enhance Text Retrieval with Transfer Learning",
      "source_url": "https://www.salesforce.com/blog/sfr-embedding/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/01#sfr-embedding-mistral"
    },
    {
      "slug": "text-embedding-3",
      "date": "2024-01-25",
      "model": "text-embedding-3-small and text-embedding-3-large",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "embedding",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "embedding"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "OpenAI released two new embedding models, text-embedding-3-small and text-embedding-3-large, succeeding text-embedding-ada-002 and adding a dimensions parameter that shortens embeddings while keeping their concept-representing properties.",
      "highlights": [
        "text-embedding-3-large creates embeddings with up to 3072 dimensions",
        "Developers can shorten embeddings natively via the dimensions API parameter",
        "Announced alongside gpt-4-0125-preview, the gpt-4-turbo-preview alias, and text-moderation-007"
      ],
      "source_title": "New embedding models and API updates",
      "source_url": "https://openai.com/index/new-embedding-models-and-api-updates/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/01#text-embedding-3"
    },
    {
      "slug": "niji-6",
      "date": "2024-01-24",
      "model": "Niji 6",
      "lab": "Midjourney",
      "lab_slug": "midjourney",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Anime and illustration model from the Niji series developed with Spellbrush, following detailed prompts more closely, writing simple text placed in quotes and adding --style raw for less anime styling.",
      "highlights": [
        "Developed in collaboration with Spellbrush",
        "A June 7, 2024 update improved rendering of Japanese kana and simple Chinese characters",
        "Supports Character Reference"
      ],
      "source_title": "Welcome to Niji V6!",
      "source_url": "https://nijijourney.com/blog/niji-updates-welcome-to-niji-v6",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/01#niji-6"
    },
    {
      "slug": "yi-vl",
      "date": "2024-01-23",
      "model": "Yi-VL",
      "lab": "01.AI",
      "lab_slug": "01-ai",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "6B, 34B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Yi-VL is 01.AI's open bilingual vision-language model family in 6B and 34B sizes, pairing a CLIP ViT-H/14 vision encoder with Yi language models for English and Chinese multimodal chat.",
      "highlights": [
        "Image understanding at 448 by 448 resolution",
        "Supports English and Chinese conversations, including text recognition in images",
        "Trained in three stages covering the ViT, projection module, and full model"
      ],
      "source_title": "GitHub - 01-ai/Yi: A series of large language models trained from scratch by developers @01-ai",
      "source_url": "https://github.com/01-ai/Yi",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/01#yi-vl"
    },
    {
      "slug": "stable-lm-2-1-6b",
      "date": "2024-01-19",
      "model": "Stable LM 2 1.6B",
      "lab": "Stability AI",
      "lab_slug": "stability-ai",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "1.6B",
      "model_license": "Stability AI Community License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "First model of the Stable LM 2 series, a 1.6B-parameter small language model with base and instruction-tuned versions trained on about 2 trillion tokens in seven languages.",
      "highlights": [
        "Languages: English, Spanish, German, Italian, French, Portuguese and Dutch",
        "Commercial use through a Stability AI Membership",
        "Updated version with tool use released April 8, 2024"
      ],
      "source_title": "Introducing Stable LM 2 1.6B",
      "source_url": "https://stability.ai/news-updates/introducing-stable-lm-2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/01#stable-lm-2-1-6b"
    },
    {
      "slug": "qwen-vl-max",
      "date": "2024-01-18",
      "model": "Qwen-VL-Max",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Qwen-VL-Max is Qwen's most capable vision-language model at launch, an upgrade over Qwen-VL-Plus with better detailed and text recognition. It supports images up to millions of pixels at arbitrary aspect ratios.",
      "highlights": [
        "Accessible through Hugging Face Spaces, the Qwen website, and DashScope APIs",
        "Upgraded alongside Qwen-VL-Plus",
        "Proprietary successor to the open Qwen-VL"
      ],
      "source_title": "Qwen-VL (GitHub)",
      "source_url": "https://github.com/QwenLM/Qwen-VL",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/01#qwen-vl-max"
    },
    {
      "slug": "stable-code-3b",
      "date": "2024-01-16",
      "model": "Stable Code 3B",
      "lab": "Stability AI",
      "lab_slug": "stability-ai",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 16384,
      "parameters": "3B",
      "model_license": "Stability AI Community License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "3B-parameter code completion model built on Stable LM 3B and trained on 18 programming languages, adding fill-in-the-middle support and training on sequences up to 16,384 tokens.",
      "highlights": [
        "Fill-in-the-middle code completion",
        "Trained on 18 programming languages",
        "Included in the Stability AI Membership for commercial use"
      ],
      "source_title": "Stable Code 3B: Coding on the Edge",
      "source_url": "https://stability.ai/news-updates/stable-code-2024-llm-code-completion-release",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/01#stable-code-3b"
    },
    {
      "slug": "nous-hermes-2-mixtral",
      "date": "2024-01-15",
      "model": "Nous-Hermes-2-Mixtral-8x7B",
      "lab": "Nous Research",
      "lab_slug": "nous-research",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Hermes 2 fine-tune of Mistral AI's Mixtral 8x7B mixture-of-experts model from Nous Research, released in supervised fine-tuned and DPO-trained versions on Hugging Face.",
      "highlights": [
        "SFT and DPO versions released together",
        "Quantized GGUF version published the same day",
        "Built on Mixtral 8x7B"
      ],
      "source_title": "Releases | Nous Research",
      "source_url": "https://nousresearch.com/releases",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/01#nous-hermes-2-mixtral"
    },
    {
      "slug": "deepseekmoe-16b",
      "date": "2024-01-11",
      "model": "DeepSeekMoE 16B",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "MoE: 16.4B total",
      "model_license": "DeepSeek Model License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "DeepSeekMoE 16B is a mixture-of-experts language model that uses fine-grained expert segmentation and shared expert isolation. Trained on 2T English and Chinese tokens, it matches DeepSeek 7B with about 40% of the computation.",
      "highlights": [
        "16.4B total parameters",
        "Deployable on a single 40GB GPU without quantization",
        "Released as Base and Chat models"
      ],
      "source_title": "DeepSeekMoE: Towards Ultimate Expert Specialization in Mixture-of-Experts Language Models (GitHub)",
      "source_url": "https://github.com/deepseek-ai/DeepSeek-MoE",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2024/01#deepseekmoe-16b"
    },
    {
      "slug": "nous-hermes-2-yi-34b",
      "date": "2023-12-25",
      "model": "Nous-Hermes-2-Yi-34B",
      "lab": "Nous Research",
      "lab_slug": "nous-research",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "34B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Hermes 2 model from Nous Research, a fine-tune of the Yi 34B base model, released openly on Hugging Face under Apache 2.0 in December 2023.",
      "highlights": [
        "Listed by Nous Research as Hermes on Yi",
        "Built on the Yi 34B base model",
        "Published under Apache 2.0"
      ],
      "source_title": "Releases | Nous Research",
      "source_url": "https://nousresearch.com/releases",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/12#nous-hermes-2-yi-34b"
    },
    {
      "slug": "midjourney-v6",
      "date": "2023-12-20",
      "model": "Midjourney V6",
      "lab": "Midjourney",
      "lab_slug": "midjourney",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Sixth-generation Midjourney image model with improved prompt accuracy for longer inputs, better coherence and knowledge, and advanced image prompting and remixing.",
      "highlights": [
        "Default model from February 14 to July 30, 2024",
        "Supports Character Reference with --cref",
        "Released in December 2023, ahead of becoming the default in February 2024"
      ],
      "source_title": "Version - Midjourney",
      "source_url": "https://docs.midjourney.com/hc/en-us/articles/32199405667853-Version",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/12#midjourney-v6"
    },
    {
      "slug": "cogagent",
      "date": "2023-12-15",
      "model": "CogAgent",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "18B",
      "model_license": "CogVLM License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "CogAgent is Z.ai's open visual language model for GUI agents, improved from CogVLM. CogAgent-18B pairs 11B visual parameters with 7B language parameters and understands images at 1120 by 1120 resolution.",
      "highlights": [
        "Adds GUI agent abilities on top of CogVLM",
        "Supports high-resolution 1120 by 1120 image understanding",
        "Accepted as a CVPR 2024 Highlight"
      ],
      "source_title": "CogVLM & CogAgent (GitHub)",
      "source_url": "https://github.com/zai-org/CogVLM",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/12#cogagent"
    },
    {
      "slug": "imagen-2",
      "date": "2023-12-13",
      "model": "Imagen 2",
      "lab": "Google",
      "lab_slug": "google",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Cloud platforms"
      ],
      "summary": "Imagen 2 is Google's second-generation text-to-image model, launched on Vertex AI with improved photorealism, text rendering in images and logo generation. It also added long-form captioning and visual question answering on images.",
      "highlights": [
        "Generally available on Vertex AI for allowlisted customers",
        "Prompts in six additional languages in preview",
        "Integrated with SynthID digital watermarking for allowlisted customers"
      ],
      "source_title": "Imagen 2 on Vertex AI is now generally available",
      "source_url": "https://cloud.google.com/blog/products/ai-machine-learning/imagen-2-on-vertex-ai-is-now-generally-available",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/12#imagen-2"
    },
    {
      "slug": "phi-2",
      "date": "2023-12-12",
      "model": "Phi-2",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 2048,
      "parameters": "2.7B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "Phi-2 is a 2.7B Microsoft Research base language model trained on 1.4T tokens of synthetic textbook-quality and filtered web data, scaled up from Phi-1.5 and offered as a platform for research.",
      "highlights": [
        "Base model without RLHF or instruction tuning",
        "Trained for 14 days on 96 A100 GPUs",
        "Released in the Azure AI Studio model catalog"
      ],
      "source_title": "Phi-2: The surprising power of small language models",
      "source_url": "https://www.microsoft.com/en-us/research/blog/phi-2-the-surprising-power-of-small-language-models/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/12#phi-2"
    },
    {
      "slug": "mistral-7b-v0-2",
      "date": "2023-12-11",
      "model": "Mistral 7B v0.2",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32000,
      "parameters": null,
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Minor release of Mistral 7B whose instruct version served the mistral-tiny endpoint at the beta launch of la Plateforme, with the context window extended to 32K tokens.",
      "highlights": [
        "Context window extended from 8K to 32K tokens",
        "Mistral 7B Instruct v0.2 served the mistral-tiny endpoint, later renamed open-mistral-7b",
        "Released under Apache 2.0"
      ],
      "source_title": "Mistral 7B v0.2 | Mistral Docs",
      "source_url": "https://docs.mistral.ai/models/mistral-7b-0-2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/12#mistral-7b-v0-2"
    },
    {
      "slug": "mistral-embed",
      "date": "2023-12-11",
      "model": "Mistral Embed",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "embedding",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "embedding"
      ],
      "context_window": 8000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "General-purpose text embedding model launched as the mistral-embed endpoint on la Plateforme, producing 1024-dimensional vectors designed with retrieval in mind.",
      "highlights": [
        "Outputs 1024-dimensional embeddings",
        "Launched alongside the first la Plateforme chat endpoints",
        "API model ID mistral-embed-2312"
      ],
      "source_title": "La Plateforme | Mistral AI",
      "source_url": "https://mistral.ai/news/la-plateforme",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/12#mistral-embed"
    },
    {
      "slug": "mistral-medium-2312",
      "date": "2023-12-11",
      "model": "Mistral Medium",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Mistral AI's first proprietary model, a prototype served as the highest-quality mistral-medium endpoint when la Plateforme opened in beta with three chat endpoints and an embedding endpoint.",
      "highlights": [
        "Launched in beta with the first la Plateforme endpoints",
        "Covers English, French, Italian, German, Spanish and code",
        "Later dated as mistral-medium-2312"
      ],
      "source_title": "La Plateforme | Mistral AI",
      "source_url": "https://mistral.ai/news/la-plateforme",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/12#mistral-medium-2312"
    },
    {
      "slug": "mixtral-8x7b",
      "date": "2023-12-11",
      "model": "Mixtral 8x7B",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32000,
      "parameters": "MoE: 46.7B total, 12.9B active",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Sparse mixture-of-experts model in which a router picks two of eight expert groups per token, released with open weights under Apache 2.0 alongside an instruct version.",
      "highlights": [
        "Handles English, French, Italian, German and Spanish",
        "Served the original mistral-small endpoint on la Plateforme",
        "Uses 12.9B of its 46.7B parameters per token"
      ],
      "source_title": "Mixtral of experts | Mistral AI",
      "source_url": "https://mistral.ai/news/mixtral-of-experts",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/12#mixtral-8x7b"
    },
    {
      "slug": "llama-guard",
      "date": "2023-12-07",
      "model": "Llama Guard",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "7B",
      "model_license": "Llama 2 Community License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Llama Guard is a 7B Llama 2 based safety classifier for filtering LLM inputs and outputs, released as part of Meta's Purple Llama trust and safety project.",
      "highlights": [
        "Instruction-tuned from Llama 2 7B",
        "Covers six harm categories in its original taxonomy",
        "Launched with CyberSec Eval cybersecurity benchmarks under Purple Llama"
      ],
      "source_title": "Announcing Purple Llama: Towards open trust and safety in the new world of generative AI",
      "source_url": "https://ai.meta.com/blog/purple-llama-open-trust-safety-generative-ai/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/12#llama-guard"
    },
    {
      "slug": "stablelm-zephyr-3b",
      "date": "2023-12-07",
      "model": "StableLM Zephyr 3B",
      "lab": "Stability AI",
      "lab_slug": "stability-ai",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "3B",
      "model_license": "Stability AI Community License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "3B-parameter chat model extending Stable LM 3B-4e1t, preference-tuned with a Zephyr-style pipeline for instruction following and question answering on edge devices.",
      "highlights": [
        "Supervised fine-tuning followed by preference tuning",
        "Launched under a non-commercial license",
        "Built on Stable LM 3B-4e1t"
      ],
      "source_title": "Introducing Stable LM Zephyr 3B: A New Addition to Stable LM, Bringing Powerful LLM Assistants to Edge Devices",
      "source_url": "https://stability.ai/news-updates/stablelm-zephyr-3b-stability-llm",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/12#stablelm-zephyr-3b"
    },
    {
      "slug": "gemini-1-0-nano",
      "date": "2023-12-06",
      "model": "Gemini 1.0 Nano",
      "lab": "Google",
      "lab_slug": "google",
      "type": "small",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Gemini 1.0 Nano was the most efficient model in Google's first Gemini generation, designed for on-device tasks. Pixel 8 Pro was the first smartphone engineered to run it.",
      "highlights": [
        "Powered Summarize in the Recorder app and Smart Reply in Gboard on Pixel 8 Pro",
        "Offered to Android developers through AICore in Android 14, starting with an early preview on Pixel 8 Pro",
        "One of three Gemini 1.0 sizes alongside Ultra and Pro"
      ],
      "source_title": "Introducing Gemini: Google's most capable AI model yet",
      "source_url": "https://blog.google/innovation-and-ai/technology/ai/google-gemini-ai/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/12#gemini-1-0-nano"
    },
    {
      "slug": "gemini-1-0-pro",
      "date": "2023-12-06",
      "model": "Gemini 1.0 Pro",
      "lab": "Google",
      "lab_slug": "google",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32768,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms",
        "Consumer app"
      ],
      "summary": "Gemini 1.0 Pro was the mid-sized model in Google's first Gemini generation, built for scaling across a wide range of tasks. A fine-tuned version began powering Bard on Dec 6, 2023, and developers gained API access on Dec 13, 2023.",
      "highlights": [
        "Bard switched to a fine-tuned Gemini Pro on Dec 6, 2023 in English in more than 170 countries and territories",
        "Gemini API and Vertex AI access began Dec 13, 2023 with a 32K context window for text",
        "A separate Gemini Pro Vision endpoint accepted text and images with text output",
        "Supported 38 languages at API launch"
      ],
      "source_title": "Introducing Gemini: Google's most capable AI model yet",
      "source_url": "https://blog.google/innovation-and-ai/technology/ai/google-gemini-ai/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/12#gemini-1-0-pro"
    },
    {
      "slug": "bytedance-skylark2",
      "date": "2023-12-05",
      "model": "Skylark2",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Skylark2 was the second generation of ByteDance's Skylark language models on Volcano Engine's Ark platform, launched as Skylark2-pro-4k and followed by 32K, lite, role-play, and turbo variants before Doubao models replaced it.",
      "highlights": [
        "Skylark2-pro-4k V1.0 released December 5, 2023",
        "Skylark2-pro-32k V1.0 followed on December 25, 2023",
        "Retired on August 20, 2024 in favor of Doubao models"
      ],
      "source_title": "模型发布公告 2024 - 火山方舟 - 火山引擎",
      "source_url": "https://docs.volcengine.com/docs/ark/model-release-announcement-2024",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/12#bytedance-skylark2"
    },
    {
      "slug": "qwen-72b",
      "date": "2023-11-30",
      "model": "Qwen-72B and Qwen-1.8B",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32000,
      "parameters": "1.8B, 72B",
      "model_license": "Tongyi Qianwen License Agreement (1.8B uses the Tongyi Qianwen Research License Agreement)",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Qwen-72B is the largest first-generation Qwen model, trained on 3T tokens with 32K context, released with the compact Qwen-1.8B and chat versions of both.",
      "highlights": [
        "Qwen-72B trained on 3.0T tokens; Qwen-1.8B on 2.2T",
        "Both support 32K context",
        "Released on ModelScope and Hugging Face"
      ],
      "source_title": "Qwen (GitHub)",
      "source_url": "https://github.com/QwenLM/Qwen",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/11#qwen-72b"
    },
    {
      "slug": "qwen-audio",
      "date": "2023-11-30",
      "model": "Qwen-Audio",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Tongyi Qianwen License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Qwen-Audio is Qwen's audio-language model that accepts diverse audio and text input, released with the instruction-tuned Qwen-Audio-Chat. Its technical paper was published on November 15, 2023.",
      "highlights": [
        "Checkpoints for Qwen-Audio and Qwen-Audio-Chat on ModelScope and Hugging Face",
        "Commercial use allowed under its license",
        "Followed by Qwen2-Audio in August 2024"
      ],
      "source_title": "Qwen-Audio (GitHub)",
      "source_url": "https://github.com/QwenLM/Qwen-Audio",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/11#qwen-audio"
    },
    {
      "slug": "meta-seamless",
      "date": "2023-11-30",
      "model": "Seamless (SeamlessM4T v2, SeamlessExpressive, SeamlessStreaming)",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": null,
      "parameters": "2.3B (SeamlessM4T v2 Large), 2.5B (SeamlessStreaming)",
      "model_license": "CC-BY-NC-4.0 (SeamlessM4T v2, SeamlessStreaming); Seamless License (SeamlessExpressive)",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Seamless is a suite of four translation models built on SeamlessM4T v2, adding expression-preserving speech translation and streaming translation with about two seconds of latency.",
      "highlights": [
        "SeamlessExpressive preserves expression across English, Spanish, German, French, Italian and Chinese",
        "SeamlessStreaming supports speech recognition and speech-to-text translation for nearly 100 languages",
        "All four models released publicly for researchers"
      ],
      "source_title": "Introducing a suite of AI language translation models that preserve expression and improve streaming",
      "source_url": "https://ai.meta.com/blog/seamless-communication/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/11#meta-seamless"
    },
    {
      "slug": "amazon-titan-image-generator-g1",
      "date": "2023-11-29",
      "model": "Amazon Titan Image Generator G1",
      "lab": "Amazon",
      "lab_slug": "amazon",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Amazon Titan Image Generator G1 is Amazon's first image generation model, launched in public preview with text-to-image generation from English prompts, text-prompted editing, inpainting and outpainting.",
      "highlights": [
        "Launched in public preview in US East (N. Virginia) and US West (Oregon)",
        "All generated images contain an invisible watermark",
        "Supports fine-tuning on proprietary data"
      ],
      "source_title": "Amazon Titan Image Generator, Multimodal Embeddings, and Text models are now available in Amazon Bedrock",
      "source_url": "https://aws.amazon.com/blogs/aws/amazon-titan-image-generator-multimodal-embeddings-and-text-models-are-now-available-in-amazon-bedrock/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/11#amazon-titan-image-generator-g1"
    },
    {
      "slug": "amazon-titan-multimodal-embeddings-g1",
      "date": "2023-11-29",
      "model": "Amazon Titan Multimodal Embeddings G1",
      "lab": "Amazon",
      "lab_slug": "amazon",
      "type": "embedding",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "embedding"
      ],
      "context_window": 128,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Amazon Titan Multimodal Embeddings G1 generates embeddings from text, images or both together, enabling multimodal search and recommendation applications in Amazon Bedrock.",
      "highlights": [
        "Accepts text, image, or text and image together, with text limited to 128 tokens",
        "Defaults to 1,024-dimension vectors, with smaller sizes available",
        "Generally available in US East (N. Virginia) and US West (Oregon)"
      ],
      "source_title": "Amazon Titan Multimodal Embeddings foundation model now generally available in Amazon Bedrock",
      "source_url": "https://aws.amazon.com/about-aws/whats-new/2023/11/amazon-titan-multimodal-embeddings-model-bedrock/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/11#amazon-titan-multimodal-embeddings-g1"
    },
    {
      "slug": "amazon-titan-text-express",
      "date": "2023-11-29",
      "model": "Amazon Titan Text G1 Express",
      "lab": "Amazon",
      "lab_slug": "amazon",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 8192,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Amazon Titan Text G1 Express is Amazon's general-purpose text generation model for open-ended generation, chat and retrieval-augmented generation, generally available in Amazon Bedrock after a limited preview of the Titan Text family.",
      "highlights": [
        "Maximum context length of 8,192 tokens",
        "Optimized for English, with 100+ additional languages in preview",
        "Titan Text was first previewed with select customers in April 2023"
      ],
      "source_title": "Amazon Titan Text models, Express and Lite, now generally available in Amazon Bedrock",
      "source_url": "https://aws.amazon.com/about-aws/whats-new/2023/11/amazon-titan-models-express-lite-bedrock/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/11#amazon-titan-text-express"
    },
    {
      "slug": "amazon-titan-text-lite",
      "date": "2023-11-29",
      "model": "Amazon Titan Text G1 Lite",
      "lab": "Amazon",
      "lab_slug": "amazon",
      "type": "small",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 4096,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Amazon Titan Text G1 Lite is the fastest model in Amazon's Titan Text family, a smaller, cost-effective option aimed at English summarization and copywriting tasks and well suited to fine-tuning.",
      "highlights": [
        "Maximum context length of 4,096 tokens",
        "Aimed at English-language tasks",
        "Launched in five AWS Regions including US East (N. Virginia) and Europe (Frankfurt)"
      ],
      "source_title": "Amazon Titan Text models, Express and Lite, now generally available in Amazon Bedrock",
      "source_url": "https://aws.amazon.com/about-aws/whats-new/2023/11/amazon-titan-models-express-lite-bedrock/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/11#amazon-titan-text-lite"
    },
    {
      "slug": "deepseek-llm",
      "date": "2023-11-29",
      "model": "DeepSeek LLM",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 4096,
      "parameters": "7B, 67B",
      "model_license": "DeepSeek Model License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "DeepSeek LLM is a general-purpose language model family released in 7B and 67B sizes as base and chat models. The models were trained from scratch on 2T English and Chinese tokens.",
      "highlights": [
        "Pre-trained with a 4,096-token sequence length",
        "DeepSeek also released intermediate base model checkpoints for research",
        "Supports commercial use under the DeepSeek Model License"
      ],
      "source_title": "DeepSeek LLM: Let there be answers (GitHub)",
      "source_url": "https://github.com/deepseek-ai/DeepSeek-LLM",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/11#deepseek-llm"
    },
    {
      "slug": "pplx-7b-70b-online",
      "date": "2023-11-29",
      "model": "pplx-7b-online and pplx-70b-online",
      "lab": "Perplexity",
      "lab_slug": "perplexity",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "pplx-7b-online: 7B (Mistral 7B base); pplx-70b-online: 70B (Llama 2 70B base)",
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Two online LLMs fine-tuned from Mistral 7B and Llama 2 70B to answer with snippets from Perplexity's own search index, giving API developers responses grounded in current web information.",
      "highlights": [
        "Available through pplx-api, which moved from beta to general release",
        "Also accessible in Perplexity Labs, the company's LLM playground",
        "Offline pplx-7b-chat and pplx-70b-chat models left alpha in the same release"
      ],
      "source_title": "Introducing PPLX Online LLMs",
      "source_url": "https://www.perplexity.ai/hub/blog/introducing-pplx-online-llms",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/11#pplx-7b-70b-online"
    },
    {
      "slug": "sdxl-turbo",
      "date": "2023-11-28",
      "model": "SDXL Turbo",
      "lab": "Stability AI",
      "lab_slug": "stability-ai",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Stability AI Community License",
      "available_at_launch": [
        "Open weights",
        "Consumer app"
      ],
      "summary": "Real-time text-to-image model distilled from SDXL 1.0 with Adversarial Diffusion Distillation, cutting the required sampling steps from 50 to one while keeping high fidelity.",
      "highlights": [
        "Single-step image generation",
        "Initially released under a non-commercial research license",
        "Beta real-time demo on Clipdrop"
      ],
      "source_title": "Introducing SDXL Turbo: A Real-Time Text-to-Image Generation Model",
      "source_url": "https://stability.ai/news-updates/stability-ai-sdxl-turbo",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/11#sdxl-turbo"
    },
    {
      "slug": "yi-34b-chat",
      "date": "2023-11-23",
      "model": "Yi-34B-Chat and Yi-6B-Chat",
      "lab": "01.AI",
      "lab_slug": "01-ai",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 4096,
      "parameters": "6B, 34B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "01.AI open-sourced Yi-34B-Chat and Yi-6B-Chat, chat models fine-tuned from its Yi base models, together with 4-bit AWQ and 8-bit GPTQ quantized versions for consumer-grade GPUs.",
      "highlights": [
        "Chat models trained with supervised fine-tuning",
        "Released with four quantized variants",
        "Yi Series Models Community License updated to v2.1 the same day; weights are now under Apache 2.0"
      ],
      "source_title": "GitHub - 01-ai/Yi: A series of large language models trained from scratch by developers @01-ai",
      "source_url": "https://github.com/01-ai/Yi",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/11#yi-34b-chat"
    },
    {
      "slug": "elevenlabs-speech-to-speech",
      "date": "2023-11-22",
      "model": "Speech to Speech (Voice Changer)",
      "lab": "ElevenLabs",
      "lab_slug": "elevenlabs",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Voice conversion model, originally called speech-to-speech, that makes a recording sound as if spoken by another voice while preserving emotion, timing, pacing and pronunciation.",
      "highlights": [
        "Preserves the original performance",
        "Added to Speech Synthesis",
        "Multilingual version eleven_multilingual_sts_v2 followed"
      ],
      "source_title": "Speech to Speech",
      "source_url": "https://elevenlabs.io/blog/speech-to-speech",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/11#elevenlabs-speech-to-speech"
    },
    {
      "slug": "claude-2-1",
      "date": "2023-11-21",
      "model": "Claude 2.1",
      "lab": "Anthropic",
      "lab_slug": "anthropic",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 200000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Claude 2.1 doubled Claude's context window to 200,000 tokens and, according to Anthropic, made fewer false statements than Claude 2.0. It also added system prompts and a beta tool use feature.",
      "highlights": [
        "API model ID claude-2.1",
        "200,000-token context window",
        "Powers claude.ai free and Pro tiers; 200K context use on claude.ai reserved for Claude Pro"
      ],
      "source_title": "Introducing Claude 2.1",
      "source_url": "https://www.anthropic.com/news/claude-2-1",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/11#claude-2-1"
    },
    {
      "slug": "stable-video-diffusion",
      "date": "2023-11-21",
      "model": "Stable Video Diffusion",
      "lab": "Stability AI",
      "lab_slug": "stability-ai",
      "type": "video",
      "weights": "open-weights",
      "input_modalities": [
        "image"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Stability AI Community License",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "Stability AI's first open video model, released in research preview as two image-to-video models generating 14 and 25 frames at customizable frame rates between 3 and 30 frames per second.",
      "highlights": [
        "Code on GitHub and weights on Hugging Face",
        "API access added on December 20, 2023",
        "Current Hugging Face license is the Stability AI Community License"
      ],
      "source_title": "Introducing Stable Video Diffusion",
      "source_url": "https://stability.ai/news-updates/stable-video-diffusion-open-ai-video-model",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/11#stable-video-diffusion"
    },
    {
      "slug": "orca-2",
      "date": "2023-11-20",
      "model": "Orca 2",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "7B, 13B",
      "model_license": "Microsoft Research License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Orca 2 is a pair of Microsoft Research models fine-tuned from Llama 2 on synthetic data to teach smaller models different reasoning strategies, such as step-by-step processing and direct answering.",
      "highlights": [
        "Released in 7B and 13B sizes",
        "Fine-tuned from Llama 2 base models",
        "Not trained with RLHF for safety, per Microsoft Research"
      ],
      "source_title": "Orca 2: Teaching Small Language Models How to Reason",
      "source_url": "https://www.microsoft.com/en-us/research/blog/orca-2-teaching-small-language-models-how-to-reason/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/11#orca-2"
    },
    {
      "slug": "lyria",
      "date": "2023-11-16",
      "model": "Lyria",
      "lab": "Google",
      "lab_slug": "google",
      "type": "music",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Lyria is Google DeepMind's music generation model, announced with YouTube, built to generate music with instrumentals and vocals, perform transformation and continuation tasks and give finer control over style and performance.",
      "highlights": [
        "Trialed in Dream Track, a YouTube Shorts experiment for a limited set of creators",
        "Dream Track generated 30-second soundtracks with lyrics, backing track and an AI voice in a participating artist's style",
        "All content published by Lyria was watermarked with SynthID"
      ],
      "source_title": "Transforming the future of music creation",
      "source_url": "https://deepmind.google/blog/transforming-the-future-of-music-creation/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/11#lyria"
    },
    {
      "slug": "nemotron-3-8b",
      "date": "2023-11-15",
      "model": "Nemotron-3 8B",
      "lab": "NVIDIA",
      "lab_slug": "nvidia",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "8B",
      "model_license": null,
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "Family of 8B enterprise language models released as NVIDIA AI Foundation Models, with base, chat and question-answering variants for building custom chatbots and copilots.",
      "highlights": [
        "Chat variants tuned with SFT, RLHF and SteerLM",
        "Available in the Azure AI model catalog, Hugging Face and NVIDIA's model hub",
        "Includes a Nemotron-3-8B-QA model"
      ],
      "source_title": "NVIDIA AI Foundation Models: Build Custom Enterprise Chatbots and Co-Pilots with Production-Ready LLMs | NVIDIA Technical Blog",
      "source_url": "https://developer.nvidia.com/blog/nvidia-ai-foundation-models-build-custom-enterprise-chatbots-and-co-pilots-with-production-ready-llms/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/11#nemotron-3-8b"
    },
    {
      "slug": "eleven-turbo-v2",
      "date": "2023-11-09",
      "model": "Eleven Turbo v2",
      "lab": "ElevenLabs",
      "lab_slug": "elevenlabs",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Low-latency English speech model generating speech at about 400ms latency, over twice as fast as ElevenLabs' V1 models with quality on par with Multilingual v2.",
      "highlights": [
        "About 400ms latency",
        "Added mulaw 8kHz output for VoIP",
        "Later outclassed by Flash v2"
      ],
      "source_title": "Turbo v2 is here",
      "source_url": "https://elevenlabs.io/blog/turbo-v2-is-here",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/11#eleven-turbo-v2"
    },
    {
      "slug": "gpt-3-5-turbo-1106",
      "date": "2023-11-06",
      "model": "gpt-3.5-turbo-1106",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 16385,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "An updated GPT-3.5 Turbo model announced at OpenAI DevDay, supporting a 16K context window by default along with improved instruction following, JSON mode, and parallel function calling.",
      "highlights": [
        "API model ID gpt-3.5-turbo-1106",
        "16,385-token context window per the GPT-3.5 Turbo docs model page",
        "Fine-tuning supports 16K context with the new model",
        "Older gpt-3.5-turbo-0613 stayed accessible until June 13, 2024"
      ],
      "source_title": "New models and developer products announced at DevDay",
      "source_url": "https://openai.com/index/new-models-and-developer-products-announced-at-devday/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/11#gpt-3-5-turbo-1106"
    },
    {
      "slug": "gpt-4-turbo",
      "date": "2023-11-06",
      "model": "GPT-4 Turbo and GPT-4 Turbo with Vision",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 128000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "GPT-4 Turbo is a preview of the next generation of GPT-4 with a 128K context window, knowledge of events up to April 2023, JSON mode, and improved function calling, plus a vision variant that accepts image inputs.",
      "highlights": [
        "Preview model IDs gpt-4-1106-preview and gpt-4-vision-preview",
        "Supports parallel function calling, JSON mode, and the seed parameter",
        "Generally available with vision on Apr 9, 2024 as gpt-4-turbo-2024-04-09 (API changelog)",
        "Updated preview gpt-4-0125-preview followed on Jan 25, 2024"
      ],
      "source_title": "New models and developer products announced at DevDay",
      "source_url": "https://openai.com/index/new-models-and-developer-products-announced-at-devday/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/11#gpt-4-turbo"
    },
    {
      "slug": "tts-1",
      "date": "2023-11-06",
      "model": "TTS (tts-1 and tts-1-hd)",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "OpenAI launched a text-to-speech API with a new TTS model offered in two variants, tts-1 optimized for real-time use and tts-1-hd optimized for quality, with six preset voices to choose from.",
      "highlights": [
        "Model IDs tts-1 and tts-1-hd in the Audio API speech endpoint",
        "Six preset voices at launch",
        "Announced at OpenAI DevDay alongside GPT-4 Turbo and the DALL·E 3 API"
      ],
      "source_title": "New models and developer products announced at DevDay",
      "source_url": "https://openai.com/index/new-models-and-developer-products-announced-at-devday/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/11#tts-1"
    },
    {
      "slug": "whisper-large-v3",
      "date": "2023-11-06",
      "model": "Whisper large-v3",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "1550M",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Whisper large-v3 is the next version of OpenAI's open-source automatic speech recognition model, released at DevDay with improved performance across languages, a 128-bin mel spectrogram input, and a new Cantonese language token.",
      "highlights": [
        "Released via the openai/whisper GitHub repo and on Hugging Face as openai/whisper-large-v3",
        "Same architecture as large and large-v2, with 1550M parameters",
        "Trained on 1 million hours of weakly labeled and 4 million hours of pseudo-labeled audio",
        "GitHub README states MIT for code and weights; the Hugging Face card is tagged apache-2.0"
      ],
      "source_title": "New models and developer products announced at DevDay",
      "source_url": "https://openai.com/index/new-models-and-developer-products-announced-at-devday/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/11#whisper-large-v3"
    },
    {
      "slug": "yi-34b-200k",
      "date": "2023-11-05",
      "model": "Yi-6B-200K and Yi-34B-200K",
      "lab": "01.AI",
      "lab_slug": "01-ai",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 200000,
      "parameters": "6B, 34B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "01.AI released Yi-6B-200K and Yi-34B-200K, base models with the same sizes as the original Yi release but a context window extended to 200K tokens.",
      "highlights": [
        "200K tokens is roughly 400,000 Chinese characters, per 01.AI",
        "Yi-34B-200K long-text capability was enhanced on March 7, 2024 with 5B tokens of long-context pre-training",
        "Bilingual English and Chinese base models"
      ],
      "source_title": "GitHub - 01-ai/Yi: A series of large language models trained from scratch by developers @01-ai",
      "source_url": "https://github.com/01-ai/Yi",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/11#yi-34b-200k"
    },
    {
      "slug": "grok-1",
      "date": "2023-11-03",
      "model": "Grok (Grok-1)",
      "lab": "xAI",
      "lab_slug": "xai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 8192,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Grok is xAI's first chatbot, powered by the Grok-1 language model developed over four months, with real-time knowledge through the X platform. It launched as an early beta for a limited number of US users.",
      "highlights": [
        "Grok-1 model card lists an 8,192-token context length",
        "Succeeded the 33B-parameter Grok-0 prototype",
        "Offered first via a waitlist; company timeline lists Grok on X on December 7, 2023"
      ],
      "source_title": "Announcing Grok | SpaceXSI",
      "source_url": "https://x.ai/news/grok",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/11#grok-1"
    },
    {
      "slug": "deepseek-coder",
      "date": "2023-11-02",
      "model": "DeepSeek Coder",
      "lab": "DeepSeek",
      "lab_slug": "deepseek",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 16000,
      "parameters": "1.3B, 5.7B, 6.7B, 33B",
      "model_license": "DeepSeek Model License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "DeepSeek Coder is a series of open-weights code models trained from scratch on 2T tokens of 87% code and 13% natural language. Base models support project-level code completion and infilling.",
      "highlights": [
        "Sizes of 1.3B, 5.7B, 6.7B, and 33B, with Base and Instruct models",
        "Pre-trained with a 16K window and a fill-in-the-blank task",
        "Covers more than 80 programming languages; supports commercial use"
      ],
      "source_title": "DeepSeek Coder: Let the Code Write Itself (GitHub)",
      "source_url": "https://github.com/deepseek-ai/DeepSeek-Coder",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/11#deepseek-coder"
    },
    {
      "slug": "cohere-embed-v3",
      "date": "2023-11-02",
      "model": "Embed v3",
      "lab": "Cohere",
      "lab_slug": "cohere",
      "type": "embedding",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "embedding"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Embedding model that judges both how well a document matches a query's topic and the quality of its content, aimed at search and RAG over noisy real-world data.",
      "highlights": [
        "Uses compression-aware training",
        "English, multilingual and light variants published",
        "Later extended to image inputs as multimodal Embed 3"
      ],
      "source_title": "Introducing Embed v3",
      "source_url": "https://cohere.com/blog/introducing-embed-v3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/11#cohere-embed-v3"
    },
    {
      "slug": "yi-34b",
      "date": "2023-11-02",
      "model": "Yi-6B and Yi-34B",
      "lab": "01.AI",
      "lab_slug": "01-ai",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 4096,
      "parameters": "6B, 34B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Yi-6B and Yi-34B were 01.AI's first public release: bilingual English and Chinese base language models trained from scratch on 3T tokens.",
      "highlights": [
        "Trained with 4K sequence length and extendable to 32K at inference time",
        "Training data runs up to June 2023",
        "Originally released under the Yi Series Models Community License; weights are now distributed under Apache 2.0"
      ],
      "source_title": "GitHub - 01-ai/Yi: A series of large language models trained from scratch by developers @01-ai",
      "source_url": "https://github.com/01-ai/Yi",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/11#yi-34b"
    },
    {
      "slug": "chatglm3-6b",
      "date": "2023-10-27",
      "model": "ChatGLM3-6B",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 8000,
      "parameters": "6B",
      "model_license": "ChatGLM3-6B License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "ChatGLM3-6B is the third generation of Z.ai's open bilingual chat model series. The series includes ChatGLM3-6B-Base and the long-context variants ChatGLM3-6B-32K and ChatGLM3-6B-128K.",
      "highlights": [
        "ChatGLM3-6B and ChatGLM3-6B-Base support 8K context",
        "ChatGLM3-6B-32K supports 32K context",
        "Code under Apache 2.0; weights under the ChatGLM3-6B License"
      ],
      "source_title": "ChatGLM3 (GitHub)",
      "source_url": "https://github.com/zai-org/ChatGLM3",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/10#chatglm3-6b"
    },
    {
      "slug": "dall-e-3",
      "date": "2023-10-19",
      "model": "DALL·E 3",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "Enterprise"
      ],
      "summary": "DALL·E 3 is OpenAI's text-to-image model, built natively on ChatGPT so the assistant can write and refine detailed prompts, with better rendering of text, hands, and faces than DALL·E 2. It launched in ChatGPT Plus and Enterprise.",
      "highlights": [
        "Announced as a research preview in September 2023",
        "API access as dall-e-3 in the Images API from DevDay on Nov 6, 2023",
        "Supports landscape and portrait aspect ratios",
        "Since deprecated and removed from the API, per its docs model page"
      ],
      "source_title": "DALL·E 3 is now available in ChatGPT Plus and Enterprise",
      "source_url": "https://openai.com/index/dall-e-3-is-now-available-in-chatgpt-plus-and-enterprise/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/10#dall-e-3"
    },
    {
      "slug": "ernie-4-0",
      "date": "2023-10-17",
      "model": "ERNIE 4.0",
      "lab": "Baidu",
      "lab_slug": "baidu",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "ERNIE 4.0 is Baidu's next-generation foundation model, unveiled at Baidu World 2023 with upgrades in understanding, generation, reasoning, and memory, and used to power a new version of ERNIE Bot.",
      "highlights": [
        "Beta testing began in September 2023, before the launch",
        "Accessible to invited users on ERNIE Bot at launch",
        "Enterprise API testing offered through the Qianfan foundation model platform"
      ],
      "source_title": "Baidu Launches ERNIE 4.0 Foundation Model, Leading a New Wave of AI-Native Applications",
      "source_url": "https://www.prnewswire.com/news-releases/baidu-launches-ernie-4-0-foundation-model-leading-a-new-wave-of-ai-native-applications-301958681.html",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/10#ernie-4-0"
    },
    {
      "slug": "cogvlm",
      "date": "2023-10-05",
      "model": "CogVLM",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "17B",
      "model_license": "CogVLM License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "CogVLM is Z.ai's open visual language model for image understanding and multi-turn dialogue. CogVLM-17B pairs 10B visual parameters with 7B language parameters and supports 490 by 490 image resolution.",
      "highlights": [
        "CogVLM-17B released October 5, 2023; a bilingual version went online on October 27",
        "Hugging Face versions of cogvlm-chat and grounding models followed on November 20, 2023",
        "Weights under the CogVLM License"
      ],
      "source_title": "CogVLM & CogAgent (GitHub)",
      "source_url": "https://github.com/zai-org/CogVLM",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/10#cogvlm"
    },
    {
      "slug": "stable-lm-3b",
      "date": "2023-10-02",
      "model": "Stable LM 3B",
      "lab": "Stability AI",
      "lab_slug": "stability-ai",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "3B",
      "model_license": "CC-BY-SA-4.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Experimental 3B-parameter compact language model designed to run on portable devices, intended as an intermediate release in the Stable LM line.",
      "highlights": [
        "3B parameters, aimed at smart devices and laptops",
        "Released on Hugging Face as stablelm-3b-4e1t",
        "Base for later models such as Stable Code 3B"
      ],
      "source_title": "Introducing Stable LM 3B: Bringing Sustainable, High-Performance Language Models to Smart Devices",
      "source_url": "https://stability.ai/news-updates/stable-lm-3b-sustainable-high-performance-language-models-smart-devices",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/10#stable-lm-3b"
    },
    {
      "slug": "amazon-titan-embeddings-g1-text",
      "date": "2023-09-28",
      "model": "Amazon Titan Embeddings G1 - Text",
      "lab": "Amazon",
      "lab_slug": "amazon",
      "type": "embedding",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "embedding"
      ],
      "context_window": 8000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Amazon Titan Embeddings G1 - Text was the first generally available Titan model, a text embedding model for search, personalization, clustering and retrieval-augmented generation, launched with general availability of Amazon Bedrock.",
      "highlights": [
        "Generally available with Amazon Bedrock on September 28, 2023",
        "8K token context window",
        "Previewed with select customers after the April 13, 2023 Titan announcement"
      ],
      "source_title": "Amazon Bedrock Is Now Generally Available",
      "source_url": "https://aws.amazon.com/blogs/aws/amazon-bedrock-is-now-generally-available-build-and-scale-generative-ai-applications-with-foundation-models/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/09#amazon-titan-embeddings-g1-text"
    },
    {
      "slug": "ibm-granite-13b",
      "date": "2023-09-28",
      "model": "Granite (granite.13b)",
      "lab": "IBM",
      "lab_slug": "ibm",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "13B",
      "model_license": null,
      "available_at_launch": [
        "Enterprise"
      ],
      "summary": "IBM's first watsonx Granite foundation models, including granite.13b.instruct and granite.13b.chat, reached general availability in watsonx.ai for enterprise language and code tasks.",
      "highlights": [
        "Covered by IBM's standard contractual IP protections",
        "Delivered through the watsonx.ai platform",
        "First models in the watsonx Granite series"
      ],
      "source_title": "IBM Announces Availability of watsonx Granite Model Series, Client Protections for IBM watsonx Models",
      "source_url": "https://newsroom.ibm.com/2023-09-28-IBM-Announces-Availability-of-watsonx-Granite-Model-Series,-Client-Protections-for-IBM-watsonx-Models",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/09#ibm-granite-13b"
    },
    {
      "slug": "mistral-7b",
      "date": "2023-09-27",
      "model": "Mistral 7B",
      "lab": "Mistral AI",
      "lab_slug": "mistral",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 8000,
      "parameters": "7.3B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Mistral AI's first model, a 7.3B-parameter open-weights language model released under Apache 2.0 with a companion chat fine-tune, built for efficient inference.",
      "highlights": [
        "Uses grouped-query attention and sliding window attention",
        "Released with a reference implementation and an instruction-tuned chat variant",
        "Apache 2.0 license allows use without restrictions"
      ],
      "source_title": "Mistral 7B | Mistral AI",
      "source_url": "https://mistral.ai/news/announcing-mistral-7b",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/09#mistral-7b"
    },
    {
      "slug": "gpt-4v",
      "date": "2023-09-25",
      "model": "GPT-4 with vision (GPT-4V)",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "multimodal",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "Enterprise"
      ],
      "summary": "OpenAI began rolling out image input in ChatGPT, powered by multimodal GPT-3.5 and GPT-4 (GPT-4V), letting users discuss photos, screenshots, and documents, alongside new voice conversations on mobile.",
      "highlights": [
        "Rolled out to ChatGPT Plus and Enterprise users over two weeks",
        "Voice chat used a new text-to-speech model plus Whisper for transcription",
        "OpenAI published the GPT-4V(ision) system card the same day",
        "Vision in the API followed at DevDay via gpt-4-vision-preview"
      ],
      "source_title": "ChatGPT can now see, hear, and speak",
      "source_url": "https://openai.com/index/chatgpt-can-now-see-hear-and-speak/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/09#gpt-4v"
    },
    {
      "slug": "qwen-14b",
      "date": "2023-09-25",
      "model": "Qwen-14B",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 8000,
      "parameters": "14B",
      "model_license": "Tongyi Qianwen License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Qwen-14B is a first-generation Qwen base model released with the Qwen-14B-Chat assistant model, along with the qwen.cpp runtime and the Qwen-Agent framework.",
      "highlights": [
        "Pre-trained on 3.0T tokens",
        "8K maximum length",
        "Released on ModelScope and Hugging Face"
      ],
      "source_title": "Qwen (GitHub)",
      "source_url": "https://github.com/QwenLM/Qwen",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/09#qwen-14b"
    },
    {
      "slug": "stable-audio",
      "date": "2023-09-13",
      "model": "Stable Audio",
      "lab": "Stability AI",
      "lab_slug": "stability-ai",
      "type": "music",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Stability AI's first music and sound generation product, generating tracks from text prompts and a requested duration through a web interface with free and Pro tiers.",
      "highlights": [
        "Free tier generates tracks up to 45 seconds",
        "Pro subscription produces 90-second tracks for commercial use",
        "Users set a desired track length along with the text prompt"
      ],
      "source_title": "Announcing Stable Audio, a product for music & sound generation",
      "source_url": "https://stability.ai/news-updates/stable-audio-using-ai-to-generate-music",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/09#stable-audio"
    },
    {
      "slug": "phi-1-5",
      "date": "2023-09-11",
      "model": "phi-1.5",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "small",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "1.3B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "phi-1.5 is a 1.3B Microsoft Research language model trained on the phi-1 data sources plus synthetic NLP texts, extending the textbook-quality data approach from Python coding to common sense reasoning and general language.",
      "highlights": [
        "1.3B parameters trained on 150B tokens",
        "Excludes generic web-crawl sources such as Common Crawl",
        "Released on Hugging Face under the MIT license"
      ],
      "source_title": "microsoft/phi-1_5 on Hugging Face",
      "source_url": "https://huggingface.co/microsoft/phi-1_5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/09#phi-1-5"
    },
    {
      "slug": "code-llama",
      "date": "2023-08-24",
      "model": "Code Llama",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 100000,
      "parameters": "7B, 13B, 34B",
      "model_license": "Llama 2 Community License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Code Llama is a family of code-specialized Llama 2 models, released in base, Python and Instruct variants for code generation and completion.",
      "highlights": [
        "Stable generations with up to 100,000 tokens of context",
        "Code Llama - Python was further fine-tuned on 100B tokens of Python code",
        "Released under the same community license as Llama 2"
      ],
      "source_title": "Introducing Code Llama, a state-of-the-art large language model for coding",
      "source_url": "https://ai.meta.com/blog/code-llama-large-language-model-coding/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/08#code-llama"
    },
    {
      "slug": "davinci-002",
      "date": "2023-08-22",
      "model": "babbage-002 and davinci-002",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "babbage-002 and davinci-002 are GPT base models without instruction tuning, released as replacements for the original GPT-3 base models in the legacy Completions API and usable as base or fine-tuned models.",
      "highlights": [
        "Replace ada, babbage, curie, and davinci, which were set to be turned off on Jan 4, 2024",
        "Announced together with fine-tuning for GPT-3.5 Turbo",
        "16,384 max output tokens and Sep 1, 2021 knowledge cutoff per the docs model pages"
      ],
      "source_title": "GPT-3.5 Turbo fine-tuning and API updates",
      "source_url": "https://openai.com/index/gpt-3-5-turbo-fine-tuning-and-api-updates/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/08#davinci-002"
    },
    {
      "slug": "eleven-multilingual-v2",
      "date": "2023-08-22",
      "model": "Eleven Multilingual v2",
      "lab": "ElevenLabs",
      "lab_slug": "elevenlabs",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Foundational multilingual speech synthesis model supporting 29 languages, released as ElevenLabs officially ended its beta phase and positioned as its lifelike model with rich emotional expression.",
      "highlights": [
        "Supports 29 languages",
        "Model ID eleven_multilingual_v2",
        "Remains a flagship text-to-speech model in the API"
      ],
      "source_title": "Eleven Multilingual v2",
      "source_url": "https://elevenlabs.io/blog/eleven-multilingual-v2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/08#eleven-multilingual-v2"
    },
    {
      "slug": "idefics",
      "date": "2023-08-22",
      "model": "IDEFICS",
      "lab": "Hugging Face",
      "lab_slug": "hugging-face",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "9B and 80B",
      "model_license": null,
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Open-access visual language model from Hugging Face that reproduces DeepMind's closed Flamingo model, accepting arbitrary sequences of images and text and generating text responses.",
      "highlights": [
        "Trained partly on the new OBELICS dataset of 141 million interleaved documents",
        "Built on CLIP-ViT-H-14 and LLaMA 65B",
        "Released in 9B and 80B variants with instruct versions"
      ],
      "source_title": "Introducing IDEFICS: An Open Reproduction of State-of-the-art Visual Langage Model",
      "source_url": "https://huggingface.co/blog/idefics",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/08#idefics"
    },
    {
      "slug": "ideogram-0-1",
      "date": "2023-08-22",
      "model": "Ideogram 0.1",
      "lab": "Ideogram",
      "lab_slug": "ideogram",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Ideogram's first publicly launched text-to-image model, introduced with the announcement of Ideogram AI and opened to everyone for free one week later.",
      "highlights": [
        "Launched with the announcement of Ideogram AI",
        "Opened to everyone for free on August 29, 2023"
      ],
      "source_title": "Blog Posts | Ideogram",
      "source_url": "https://docs.ideogram.ai/about-ideogram/blog-posts",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/08#ideogram-0-1"
    },
    {
      "slug": "qwen-vl",
      "date": "2023-08-22",
      "model": "Qwen-VL",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Tongyi Qianwen License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Qwen-VL is Qwen's first open vision-language model, released with Qwen-VL-Chat. It uses 448 by 448 image resolution for fine-grained text recognition, document question answering, and bounding box annotation.",
      "highlights": [
        "448 by 448 resolution, higher than the 224 by 224 used by other open models at the time",
        "Released on ModelScope and Hugging Face with a paper",
        "Commercial use allowed under its license"
      ],
      "source_title": "Qwen-VL (GitHub)",
      "source_url": "https://github.com/QwenLM/Qwen-VL",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/08#qwen-vl"
    },
    {
      "slug": "seamlessm4t",
      "date": "2023-08-22",
      "model": "SeamlessM4T",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "audio"
      ],
      "output_modalities": [
        "text",
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "CC-BY-NC-4.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "SeamlessM4T is a single multimodal model for speech and text translation and speech recognition, covering nearly 100 input languages and 35 speech output languages plus English.",
      "highlights": [
        "Speech recognition for nearly 100 languages",
        "Speech-to-speech translation into 35 languages plus English",
        "Released publicly under CC BY-NC 4.0"
      ],
      "source_title": "Bringing the world closer together with a foundational multimodal model for speech translation",
      "source_url": "https://ai.meta.com/blog/seamless-m4t/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/08#seamlessm4t"
    },
    {
      "slug": "claude-instant-1-2",
      "date": "2023-08-09",
      "model": "Claude Instant 1.2",
      "lab": "Anthropic",
      "lab_slug": "anthropic",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Claude Instant 1.2 is an updated version of Anthropic's faster, lower-priced Claude Instant model that incorporates strengths of Claude 2, with gains in math, coding, reasoning, and safety and longer, more structured responses.",
      "highlights": [
        "API model ID claude-instant-1.2",
        "Available through the Anthropic API",
        "Improves on Claude Instant 1.1 in quote extraction, multilingual capabilities, and question answering"
      ],
      "source_title": "Releasing Claude Instant 1.2",
      "source_url": "https://www.anthropic.com/news/releasing-claude-instant-1-2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/08#claude-instant-1-2"
    },
    {
      "slug": "stablecode-alpha",
      "date": "2023-08-08",
      "model": "StableCode (Alpha)",
      "lab": "Stability AI",
      "lab_slug": "stability-ai",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "3B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Stability AI's first code models: a base model trained on BigCode's The Stack, an instruction-tuned model and a long-context completion model aimed at single and multi-line autocomplete.",
      "highlights": [
        "Three models: base, instruction-tuned and long-context completion",
        "Completion model released under Apache 2.0",
        "3B parameters"
      ],
      "source_title": "Announcing Stable Code Alpha",
      "source_url": "https://stability.ai/news-updates/stablecode-llm-generative-ai-coding",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/08#stablecode-alpha"
    },
    {
      "slug": "bytedance-skylark",
      "date": "2023-08-06",
      "model": "Skylark",
      "lab": "ByteDance Seed",
      "lab_slug": "bytedance-seed",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Skylark was ByteDance's first-generation language model family on Volcano Engine's Ark platform, released in pro and lite tiers alongside Skylark-plus and later joined by Skylark-chat, the same model used in the Doubao app.",
      "highlights": [
        "Skylark-pro and Skylark-lite V1.0 listed as new releases",
        "Skylark-chat, matching the Doubao app, added September 26, 2023",
        "Version 2.0 in October 2023 gave Skylark-lite 8K and Skylark-pro 4K context",
        "Retired on August 20, 2024"
      ],
      "source_title": "模型发布公告 2024 - 火山方舟 - 火山引擎",
      "source_url": "https://docs.volcengine.com/docs/ark/model-release-announcement-2024",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/08#bytedance-skylark"
    },
    {
      "slug": "qwen-7b",
      "date": "2023-08-03",
      "model": "Qwen-7B",
      "lab": "Alibaba Qwen",
      "lab_slug": "alibaba-qwen",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "7B",
      "model_license": "Tongyi Qianwen License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Qwen-7B is the first open model from Alibaba's Qwen (Tongyi Qianwen) team, released with the Qwen-7B-Chat assistant model on ModelScope and Hugging Face.",
      "highlights": [
        "Pre-trained on 2.4T tokens",
        "Released with a technical memo on training details",
        "Weights under the Tongyi Qianwen License Agreement"
      ],
      "source_title": "Qwen (GitHub)",
      "source_url": "https://github.com/QwenLM/Qwen",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/08#qwen-7b"
    },
    {
      "slug": "meta-audiocraft",
      "date": "2023-08-02",
      "model": "AudioCraft (MusicGen and AudioGen)",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "music",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "CC-BY-NC-4.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Meta open-sourced AudioCraft, its generative audio framework, releasing model weights and code for MusicGen text-to-music, AudioGen text-to-sound and an improved EnCodec decoder.",
      "highlights": [
        "Model weights released under CC-BY-NC 4.0 for research use",
        "Training code released under the MIT license"
      ],
      "source_title": "Open sourcing AudioCraft: Generative AI for audio made simple and available to all",
      "source_url": "https://ai.meta.com/blog/audiocraft-musicgen-audiogen-encodec-generative-ai-audio/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/08#meta-audiocraft"
    },
    {
      "slug": "chatglm2-6b-32k",
      "date": "2023-07-31",
      "model": "ChatGLM2-6B-32K",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32000,
      "parameters": "6B",
      "model_license": "ChatGLM2-6B License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "ChatGLM2-6B-32K is a long-context version of ChatGLM2-6B that improves understanding of long text. It extends the chat model's usable context to 32K tokens.",
      "highlights": [
        "Built on the ChatGLM2-6B model",
        "Targets long-document understanding",
        "Weights under the ChatGLM2-6B License"
      ],
      "source_title": "ChatGLM2-6B (GitHub)",
      "source_url": "https://github.com/zai-org/ChatGLM2-6B",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/07#chatglm2-6b-32k"
    },
    {
      "slug": "sdxl-1-0",
      "date": "2023-07-26",
      "model": "SDXL 1.0",
      "lab": "Stability AI",
      "lab_slug": "stability-ai",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "3.5B base, 6.6B ensemble",
      "model_license": "CreativeML OpenRAIL++-M",
      "available_at_launch": [
        "Open weights",
        "API",
        "Consumer app",
        "Cloud platforms"
      ],
      "summary": "Open release of Stable Diffusion XL, Stability AI's flagship image model at the time, with a 3.5B-parameter base model and a refiner forming a 6.6B-parameter pipeline at native 1024x1024.",
      "highlights": [
        "Released under the CreativeML OpenRAIL++-M License",
        "Runs on consumer GPUs with 8GB VRAM",
        "Available on Amazon SageMaker, Amazon Bedrock and DreamStudio"
      ],
      "source_title": "Announcing SDXL 1.0",
      "source_url": "https://stability.ai/news-updates/stable-diffusion-sdxl-1-announcement",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/07#sdxl-1-0"
    },
    {
      "slug": "codegeex2",
      "date": "2023-07-25",
      "model": "CodeGeeX2",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 8192,
      "parameters": "6B",
      "model_license": "CodeGeeX License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "CodeGeeX2 is the second generation of Z.ai's multilingual code generation model, built on ChatGLM2-6B with an additional 600B tokens of code pre-training. It also powers the CodeGeeX coding assistant plugins.",
      "highlights": [
        "Supports a maximum sequence length of 8,192",
        "Runs in about 6GB of GPU memory after quantization",
        "Weights open for academic research; commercial use requires registration"
      ],
      "source_title": "CodeGeeX2 (GitHub)",
      "source_url": "https://github.com/zai-org/CodeGeeX2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/07#codegeex2"
    },
    {
      "slug": "nous-hermes-llama-2",
      "date": "2023-07-21",
      "model": "Nous-Hermes-Llama2-13b",
      "lab": "Nous Research",
      "lab_slug": "nous-research",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "13B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Nous Research's Hermes fine-tune of Meta's Llama 2 13B, an early open-weights entry in the Hermes series of instruction-following models released on Hugging Face.",
      "highlights": [
        "Built on Llama 2 13B",
        "Listed by Nous Research as Hermes on Llama2",
        "Published on Hugging Face"
      ],
      "source_title": "Releases | Nous Research",
      "source_url": "https://nousresearch.com/releases",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/07#nous-hermes-llama-2"
    },
    {
      "slug": "stable-beluga",
      "date": "2023-07-21",
      "model": "Stable Beluga 1 and Stable Beluga 2",
      "lab": "Stability AI",
      "lab_slug": "stability-ai",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "65B and 70B",
      "model_license": "Stable Beluga Non-Commercial Community License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Instruction fine-tuned language models from Stability AI and its CarperAI lab, built on the LLaMA 65B and Llama 2 70B foundation models and trained on a synthetic Orca-style dataset.",
      "highlights": [
        "Released for non-commercial research use",
        "Formerly codenamed FreeWilly",
        "Trained on 600,000 synthetic instruction examples"
      ],
      "source_title": "Meet Stable Beluga 1 and Stable Beluga 2, Our Large and Mighty Instruction Fine-Tuned Language Models",
      "source_url": "https://stability.ai/news-updates/stable-beluga-large-instruction-fine-tuned-models",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/07#stable-beluga"
    },
    {
      "slug": "llama-2",
      "date": "2023-07-18",
      "model": "Llama 2",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 4096,
      "parameters": "7B, 13B, 70B",
      "model_license": "Llama 2 Community License",
      "available_at_launch": [
        "Open weights",
        "Cloud platforms"
      ],
      "summary": "Llama 2 is Meta's second-generation open model family, released free for research and commercial use with pretrained and chat fine-tuned versions in 7B, 13B and 70B sizes.",
      "highlights": [
        "Pretrained on 2 trillion tokens of publicly available data",
        "Launched with Microsoft as preferred partner, in the Azure AI model catalog",
        "Also available through AWS and Hugging Face"
      ],
      "source_title": "Meta and Microsoft Introduce the Next Generation of Llama",
      "source_url": "https://ai.meta.com/blog/llama-2/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/07#llama-2"
    },
    {
      "slug": "claude-2",
      "date": "2023-07-11",
      "model": "Claude 2",
      "lab": "Anthropic",
      "lab_slug": "anthropic",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 100000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Consumer app"
      ],
      "summary": "Claude 2 is Anthropic's second-generation model, with improved coding, math, and reasoning and longer outputs than Claude 1.3. It launched with the claude.ai public beta chat website alongside API access for businesses.",
      "highlights": [
        "API model ID claude-2.0",
        "Accepts up to 100K tokens per prompt",
        "claude.ai public beta opened to users in the US and UK"
      ],
      "source_title": "Claude 2",
      "source_url": "https://www.anthropic.com/news/claude-2",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/07#claude-2"
    },
    {
      "slug": "codegen2-5",
      "date": "2023-07-06",
      "model": "CodeGen2.5",
      "lab": "Salesforce",
      "lab_slug": "salesforce",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "7B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "7B code generation model from Salesforce Research positioned to match larger 15B-plus code models at less than half the size, with local deployment in mind.",
      "highlights": [
        "Successor to the CodeGen and CodeGen2 models",
        "Released as multi, mono and instruct variants",
        "Multi variant released under Apache 2.0"
      ],
      "source_title": "CodeGen2.5: Small, but mighty - Salesforce",
      "source_url": "https://www.salesforce.com/blog/codegen25/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/07#codegen2-5"
    },
    {
      "slug": "xgen-7b",
      "date": "2023-06-28",
      "model": "XGen-7B",
      "lab": "Salesforce",
      "lab_slug": "salesforce",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 8192,
      "parameters": "7B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Series of 7B language models trained with standard dense attention on sequences up to 8K tokens for up to 1.5 trillion tokens, plus instruction-tuned versions.",
      "highlights": [
        "Base models trained at 4K and 8K sequence lengths",
        "Instruct versions tuned on public-domain instructional data",
        "Base models released under Apache 2.0"
      ],
      "source_title": "Long Sequence Modeling with XGen: A 7B LLM Trained on 8K Input Sequence Length - Salesforce",
      "source_url": "https://www.salesforce.com/blog/xgen/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/06#xgen-7b"
    },
    {
      "slug": "chatglm2-6b",
      "date": "2023-06-25",
      "model": "ChatGLM2-6B",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 32000,
      "parameters": "6B",
      "model_license": "ChatGLM2-6B License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "ChatGLM2-6B is the second-generation version of Z.ai's open bilingual chat model, with an upgraded base model pre-trained on 1.4T Chinese and English tokens. It extends base context from 2K to 32K tokens and is trained for 8K dialogue.",
      "highlights": [
        "Uses Multi-Query Attention for faster inference and lower memory use",
        "Context length extended from 2K to 32K using FlashAttention",
        "Free commercial use allowed after registration"
      ],
      "source_title": "ChatGLM-6B (GitHub)",
      "source_url": "https://github.com/zai-org/ChatGLM-6B",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/06#chatglm2-6b"
    },
    {
      "slug": "inflection-1",
      "date": "2023-06-22",
      "model": "Inflection-1",
      "lab": "Inflection AI",
      "lab_slug": "inflection-ai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Inflection AI's in-house large language model powering Pi, trained on thousands of NVIDIA H100 GPUs and adapted through a proprietary process into a personal AI, with a technical memo detailing its evaluations.",
      "highlights": [
        "Powers Pi at pi.ai; Inflection said it would come to its conversational API",
        "Technical memo reports results for the base model before fine-tuning or alignment"
      ],
      "source_title": "Inflection-1: Pi’s Best-in-Class LLM",
      "source_url": "https://web.archive.org/web/20240618183300/https://inflection.ai/inflection-1",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/06#inflection-1"
    },
    {
      "slug": "midjourney-v5-2",
      "date": "2023-06-22",
      "model": "Midjourney V5.2",
      "lab": "Midjourney",
      "lab_slug": "midjourney",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Version 5 update producing more detailed, sharper results with better colors, contrast and composition, a slightly better understanding of prompts and stronger response to the --stylize parameter.",
      "highlights": [
        "Default model from June 22, 2023 to February 14, 2024",
        "The Discord Style Tuner was a Version 5.2 feature",
        "Midjourney's documentation lists its release as June 2023"
      ],
      "source_title": "Legacy Features - Midjourney",
      "source_url": "https://docs.midjourney.com/hc/en-us/articles/33329788681101-Legacy-Features",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/06#midjourney-v5-2"
    },
    {
      "slug": "mpt-30b",
      "date": "2023-06-22",
      "model": "MPT-30B",
      "lab": "Databricks",
      "lab_slug": "databricks",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 8192,
      "parameters": "30B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights",
        "API"
      ],
      "summary": "MosaicML Foundation Series model trained with an 8k-token context window, released by MosaicML four days before Databricks announced its agreement to acquire the company.",
      "highlights": [
        "Pretrained on 1T tokens at 2k context, then 50B tokens at 8k",
        "Partly trained on NVIDIA H100 GPUs",
        "Released with Instruct and non-commercial Chat variants"
      ],
      "source_title": "MPT-30B: Raising the bar for open-source foundation models | Databricks Blog",
      "source_url": "https://www.databricks.com/blog/mpt-30b",
      "retired_date": "2025-12-19",
      "retirement_note": "Databricks retired MPT-30B and MPT-30B Instruct from pay-per-token serving on August 30, 2024 and from provisioned throughput on December 19, 2025.",
      "retirement_source_url": "https://docs.databricks.com/aws/en/machine-learning/retired-models-policy",
      "permalink": "https://theaiengineering.com/releases/2023/06#mpt-30b"
    },
    {
      "slug": "sdxl-0-9",
      "date": "2023-06-22",
      "model": "SDXL 0.9",
      "lab": "Stability AI",
      "lab_slug": "stability-ai",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": "3.5B base, 6.6B ensemble",
      "model_license": "SDXL 0.9 Research License",
      "available_at_launch": [
        "Open weights",
        "API",
        "Consumer app"
      ],
      "summary": "Research release of Stable Diffusion XL with a 3.5B-parameter base model in a 6.6B-parameter two-model ensemble, offered on Clipdrop and the Stability API ahead of the open SDXL 1.0 release.",
      "highlights": [
        "Research weights available on application under a non-commercial, research-only license",
        "Uses two CLIP models including OpenCLIP ViT-G/14",
        "Native 1024x1024 resolution"
      ],
      "source_title": "Stability AI launches SDXL 0.9: A Leap Forward in AI Image Generation",
      "source_url": "https://stability.ai/news-updates/sdxl-09-stable-diffusion",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/06#sdxl-0-9"
    },
    {
      "slug": "phi-1",
      "date": "2023-06-20",
      "model": "phi-1",
      "lab": "Microsoft",
      "lab_slug": "microsoft",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "1.3B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "phi-1 is a 1.3B Microsoft Research model for Python code generation, trained on filtered code data and synthetic textbooks and exercises, introduced in the paper Textbooks Are All You Need.",
      "highlights": [
        "1.3B parameter Transformer specialized in Python",
        "Trained on 54B tokens including synthetic textbooks generated with GPT-3.5",
        "Weights published on Hugging Face under the MIT license"
      ],
      "source_title": "microsoft/phi-1 on Hugging Face",
      "source_url": "https://huggingface.co/microsoft/phi-1",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/06#phi-1"
    },
    {
      "slug": "gpt-3-5-turbo-0613",
      "date": "2023-06-13",
      "model": "gpt-3.5-turbo-0613 and gpt-3.5-turbo-16k",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "OpenAI released updated GPT-3.5 Turbo models: gpt-3.5-turbo-0613 adds function calling and more reliable steerability via the system message, and gpt-3.5-turbo-16k offers four times the context length of the standard 4K version.",
      "highlights": [
        "gpt-3.5-turbo-16k fits about 20 pages of text in a single request",
        "16,385-token context for gpt-3.5-turbo-16k-0613 per its docs model page",
        "Stable gpt-3.5-turbo alias was set to upgrade to the new model on June 27, 2023"
      ],
      "source_title": "Function calling and other API updates",
      "source_url": "https://openai.com/index/function-calling-and-other-api-updates/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/06#gpt-3-5-turbo-0613"
    },
    {
      "slug": "gpt-4-0613",
      "date": "2023-06-13",
      "model": "gpt-4-0613",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 8192,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "gpt-4-0613 is an updated GPT-4 snapshot that introduced function calling in the Chat Completions API, letting the model output JSON arguments for developer-described functions, alongside gpt-4-32k-0613 with the same improvements and extended context.",
      "highlights": [
        "8,192-token context for gpt-4 snapshots per the GPT-4 docs model page",
        "gpt-4-32k-0613 adds the same improvements with a longer context",
        "Stable gpt-4 and gpt-4-32k aliases were set to upgrade to the new models on June 27, 2023"
      ],
      "source_title": "Function calling and other API updates",
      "source_url": "https://openai.com/index/function-calling-and-other-api-updates/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/06#gpt-4-0613"
    },
    {
      "slug": "codet5-plus",
      "date": "2023-05-20",
      "model": "CodeT5+",
      "lab": "Salesforce",
      "lab_slug": "salesforce",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "220M to 16B",
      "model_license": "BSD-3-Clause",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Family of open code language models with improved architectures and training, using a shallow encoder and deep decoder built on frozen code LLMs for efficient scaling.",
      "highlights": [
        "Evaluated on more than 20 code benchmarks",
        "Instruction-tuned InstructCodeT5+ 16B also released",
        "Released in sizes from 220M to 16B"
      ],
      "source_title": "CodeT5+: Open Code Large Language Models",
      "source_url": "https://www.salesforce.com/blog/codet5-open-code-large-language-models/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/05#codet5-plus"
    },
    {
      "slug": "visualglm-6b",
      "date": "2023-05-17",
      "model": "VisualGLM-6B",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "VisualGLM-6B License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "VisualGLM-6B is Z.ai's open multimodal dialogue model that supports images, Chinese, and English. It builds on the 6.2B-parameter ChatGLM-6B language model and bridges vision and language with a BLIP2-Qformer.",
      "highlights": [
        "Pre-trained on 30M Chinese and 300M English image-text pairs",
        "Language model based on ChatGLM-6B",
        "Code under Apache 2.0; weights under a separate model license"
      ],
      "source_title": "VisualGLM-6B (GitHub)",
      "source_url": "https://github.com/zai-org/VisualGLM-6B",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/05#visualglm-6b"
    },
    {
      "slug": "claude-100k",
      "date": "2023-05-11",
      "model": "Claude and Claude Instant (100K context)",
      "lab": "Anthropic",
      "lab_slug": "anthropic",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 100000,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Anthropic expanded Claude's context window from 9K to 100K tokens, about 75,000 words, letting businesses submit hundreds of pages of material or long transcripts in a single prompt. The 100K versions were made available through the API.",
      "highlights": [
        "Context window expanded from 9K to 100K tokens",
        "Available in the Anthropic API",
        "Anthropic demonstrated Claude Instant reading The Great Gatsby (72K tokens) in a single prompt"
      ],
      "source_title": "Introducing 100K context windows",
      "source_url": "https://www.anthropic.com/news/100k-context-windows",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/05#claude-100k"
    },
    {
      "slug": "instructblip",
      "date": "2023-05-11",
      "model": "InstructBLIP",
      "lab": "Salesforce",
      "lab_slug": "salesforce",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Vision-language instruction-tuning framework built on BLIP-2 models, released with state-of-the-art zero-shot generalization across a wide range of vision-language tasks.",
      "highlights": [
        "Released in LAVIS with Vicuna 7B and 13B and FlanT5-XL and FlanT5-XXL variants",
        "Vicuna variants used frozen Vicuna v1.1 weights that users prepared separately"
      ],
      "source_title": "salesforce/LAVIS GitHub repository: InstructBLIP",
      "source_url": "https://github.com/salesforce/LAVIS/tree/main/projects/instructblip",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/05#instructblip"
    },
    {
      "slug": "chirp",
      "date": "2023-05-10",
      "model": "Chirp",
      "lab": "Google",
      "lab_slug": "google",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "2B",
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms"
      ],
      "summary": "Chirp is Google Cloud's speech-to-text foundation model, a version of the 2B-parameter Universal Speech Model supporting over 100 languages. It launched in preview on Vertex AI at Google I/O 2023.",
      "highlights": [
        "Trained on millions of hours of audio",
        "Available in preview to everyone with a Google Cloud account at launch",
        "Generally available through the Speech-to-Text v2 API from August 2023"
      ],
      "source_title": "Google Cloud advances generative AI at I/O: new foundation models, embeddings, and tuning tools in Vertex AI",
      "source_url": "https://cloud.google.com/blog/products/ai-machine-learning/google-cloud-launches-new-ai-models-opens-generative-ai-studio",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/05#chirp"
    },
    {
      "slug": "palm-2",
      "date": "2023-05-10",
      "model": "PaLM 2",
      "lab": "Google",
      "lab_slug": "google",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API",
        "Cloud platforms",
        "Consumer app"
      ],
      "summary": "PaLM 2 is Google's second-generation large language model, trained more heavily on multilingual text and source code, and offered in several sizes for uses ranging from mobile devices to the cloud.",
      "highlights": [
        "Offered in four sizes: Gecko, Otter, Bison and Unicorn",
        "Pre-training spanned multilingual text in more than 100 languages",
        "Opened to developers through the PaLM API and to Google Cloud customers in Vertex AI",
        "Powered Bard and over 25 products and features announced at Google I/O 2023"
      ],
      "source_title": "Google AI: What to know about the PaLM 2 large language model",
      "source_url": "https://blog.google/innovation-and-ai/products/google-palm-2-ai-large-language-model/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/05#palm-2"
    },
    {
      "slug": "mpt-7b",
      "date": "2023-05-05",
      "model": "MPT-7B",
      "lab": "Databricks",
      "lab_slug": "databricks",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "6.7B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "First MosaicML Foundation Series model, a decoder-only transformer trained from scratch on 1T tokens of text and code and licensed for commercial use, released by MosaicML before Databricks acquired the company in July 2023.",
      "highlights": [
        "Trained on the MosaicML platform in 9.5 days for about $200k",
        "StoryWriter-65k+ variant fine-tuned with a 65k-token context",
        "Instruct variant under CC BY-SA 3.0, Chat variant non-commercial"
      ],
      "source_title": "Introducing MPT-7B: A New Standard for Open-Source, Commercially Usable LLMs | Databricks Blog",
      "source_url": "https://www.databricks.com/blog/mpt-7b",
      "retired_date": "2025-12-19",
      "retirement_note": "Databricks retired MPT-7B and MPT-7B Instruct from pay-per-token serving on August 30, 2024 and from provisioned throughput on December 19, 2025.",
      "retirement_source_url": "https://docs.databricks.com/aws/en/machine-learning/retired-models-policy",
      "permalink": "https://theaiengineering.com/releases/2023/05#mpt-7b"
    },
    {
      "slug": "midjourney-v5-1",
      "date": "2023-05-04",
      "model": "Midjourney V5.1",
      "lab": "Midjourney",
      "lab_slug": "midjourney",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Update to Midjourney's version 5 model with a stronger default aesthetic for simple prompts, high coherency, more accurate natural language interpretation and fewer unwanted artifacts and borders.",
      "highlights": [
        "Supports repeating patterns with --tile",
        "Default model from May 2023 until June 22, 2023",
        "Generates 1024x1024 images"
      ],
      "source_title": "Legacy Features - Midjourney",
      "source_url": "https://docs.midjourney.com/hc/en-us/articles/33329788681101-Legacy-Features",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/05#midjourney-v5-1"
    },
    {
      "slug": "codegen2",
      "date": "2023-05-03",
      "model": "CodeGen2",
      "lab": "Salesforce",
      "lab_slug": "salesforce",
      "type": "coding",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "1B, 3.7B, 7B and 16B",
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Second generation of Salesforce's CodeGen program synthesis models, adding infill sampling and support for more programming languages than the original CodeGen family.",
      "highlights": [
        "Released in 1B, 3.7B, 7B and 16B sizes",
        "Introduced special tokens for infill sampling",
        "Released under Apache 2.0"
      ],
      "source_title": "salesforce/CodeGen GitHub repository",
      "source_url": "https://github.com/salesforce/CodeGen",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/05#codegen2"
    },
    {
      "slug": "cohere-rerank",
      "date": "2023-05-01",
      "model": "Cohere Rerank (v2.0)",
      "lab": "Cohere",
      "lab_slug": "cohere",
      "type": "reranker",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "score"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Cohere's first Rerank endpoint, a model that scores the relevance of each search result to a query so it can be added as a final stage to keyword or semantic search.",
      "highlights": [
        "Offered as rerank-english-v2.0 and rerank-multilingual-v2.0",
        "Supports more than 100 languages in the multilingual version",
        "Integrates with Elasticsearch, OpenSearch and Solr pipelines"
      ],
      "source_title": "Cohere Rerank: Say Goodbye to Irrelevant Search Results",
      "source_url": "https://cohere.com/blog/rerank",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/05#cohere-rerank"
    },
    {
      "slug": "deepfloyd-if",
      "date": "2023-04-28",
      "model": "DeepFloyd IF",
      "lab": "Stability AI",
      "lab_slug": "stability-ai",
      "type": "image",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "DeepFloyd IF License Agreement",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Cascaded pixel diffusion text-to-image model from Stability AI's DeepFloyd research lab, using a T5-XXL text encoder to render coherent text in images, released under a non-commercial research license.",
      "highlights": [
        "Research release on a non-commercial, research-permissible license",
        "Uses T5-XXL-1.1 as the text encoder",
        "Supports zero-shot image-to-image translation"
      ],
      "source_title": "Stability AI releases DeepFloyd IF, a powerful text-to-image model that can smartly integrate text into images",
      "source_url": "https://stability.ai/news-updates/deepfloyd-if-text-to-image-model",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/04#deepfloyd-if"
    },
    {
      "slug": "eleven-multilingual-v1",
      "date": "2023-04-27",
      "model": "Eleven Multilingual v1",
      "lab": "ElevenLabs",
      "lab_slug": "elevenlabs",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "Speech synthesis model building on Eleven Monolingual v1 that added seven languages: French, German, Hindi, Italian, Polish, Portuguese and Spanish.",
      "highlights": [
        "Seven new languages",
        "Available across all subscription plans",
        "Selectable in the Speech Synthesis panel"
      ],
      "source_title": "Eleven Multilingual v1",
      "source_url": "https://elevenlabs.io/blog/eleven-multilingual-v1",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/04#eleven-multilingual-v1"
    },
    {
      "slug": "suno-bark",
      "date": "2023-04-20",
      "model": "Bark",
      "lab": "Suno",
      "lab_slug": "suno",
      "type": "speech",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "audio"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Transformer-based text-to-audio model that generates realistic multilingual speech as well as music, background noise, simple sound effects and nonverbal sounds like laughing and sighing.",
      "highlights": [
        "Relicensed under MIT for commercial use on May 1, 2023",
        "Pretrained checkpoints on GitHub and Hugging Face",
        "Outputs limited to about 13 to 14 seconds per generation"
      ],
      "source_title": "suno-ai/bark: Text-Prompted Generative Audio Model",
      "source_url": "https://github.com/suno-ai/bark",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/04#suno-bark"
    },
    {
      "slug": "stablelm-alpha",
      "date": "2023-04-19",
      "model": "StableLM (Alpha)",
      "lab": "Stability AI",
      "lab_slug": "stability-ai",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 4096,
      "parameters": "3B and 7B",
      "model_license": "CC-BY-SA-4.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "First Stable LM release, 3B and 7B base language models trained on an experimental dataset built on The Pile with 1.5 trillion tokens, plus research-only instruction-tuned variants.",
      "highlights": [
        "Base models usable commercially under CC BY-SA 4.0",
        "Trained with a 4,096-token sequence length",
        "Instruction-tuned research models released alongside"
      ],
      "source_title": "Stability AI Launches the First of its Stable LM Suite of Language Models",
      "source_url": "https://stability.ai/news-updates/stability-ai-launches-the-first-of-its-stablelm-suite-of-language-models",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/04#stablelm-alpha"
    },
    {
      "slug": "databricks-dolly-2",
      "date": "2023-04-12",
      "model": "Dolly 2.0 (dolly-v2-12b)",
      "lab": "Databricks",
      "lab_slug": "databricks",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "12B",
      "model_license": "MIT",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "12B instruction-following model based on EleutherAI's Pythia family, fine-tuned only on databricks-dolly-15k, a human-written instruction dataset licensed for research and commercial use.",
      "highlights": [
        "Dataset written by more than 5,000 Databricks employees in March and April 2023",
        "Training code, dataset and weights released for commercial use",
        "databricks-dolly-15k licensed under CC BY-SA 3.0"
      ],
      "source_title": "Dolly: Open Instruction-Tuned LLM | Databricks Blog",
      "source_url": "https://www.databricks.com/blog/2023/04/12/dolly-first-open-commercially-viable-instruction-tuned-llm",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/04#databricks-dolly-2"
    },
    {
      "slug": "meta-sam",
      "date": "2023-04-05",
      "model": "Segment Anything Model (SAM)",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": "Apache-2.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Segment Anything Model is a promptable image segmentation foundation model that can segment objects from clicks or boxes and transfer zero-shot to new image domains.",
      "highlights": [
        "Released with SA-1B, over 1.1 billion masks on about 11 million images",
        "Produces a segment in about 50 milliseconds in a browser once the image embedding is computed"
      ],
      "source_title": "Introducing Segment Anything: Working toward the first foundation model for image segmentation",
      "source_url": "https://ai.meta.com/blog/segment-anything-foundation-model-image-segmentation/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/04#meta-sam"
    },
    {
      "slug": "niji-5",
      "date": "2023-04-02",
      "model": "Niji 5",
      "lab": "Midjourney",
      "lab_slug": "midjourney",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Anime and illustration model from the Niji series, built with Spellbrush and tuned for emotional, dramatic and artistic illustrations with cleaner light and shadow, precise small details and better coherency on subjects such as hands.",
      "highlights": [
        "Developed in collaboration with Spellbrush",
        "Introduced --s to tone the Niji aesthetic up or down",
        "Expressive, cute and scenic --style options followed in the weeks after launch"
      ],
      "source_title": "Introducing Niji 5!",
      "source_url": "https://nijijourney.com/blog/introducing-niji-5",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/04#niji-5"
    },
    {
      "slug": "databricks-dolly",
      "date": "2023-03-24",
      "model": "Dolly",
      "lab": "Databricks",
      "lab_slug": "databricks",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "6B",
      "model_license": "CC-BY-NC-4.0",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Databricks' first instruction-following model, made by fine-tuning a two-year-old 6B open model from EleutherAI on Alpaca data to show that instruction following does not seem to require the latest or largest models.",
      "highlights": [
        "Trained in 30 minutes on one machine",
        "Fine-tuned on instruction data from Stanford's Alpaca",
        "Released with a Databricks notebook for rebuilding it"
      ],
      "source_title": "Hello Dolly: Democratizing the magic of ChatGPT with open models",
      "source_url": "https://www.databricks.com/blog/2023/03/24/hello-dolly-democratizing-magic-chatgpt-open-models.html",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/03#databricks-dolly"
    },
    {
      "slug": "runway-gen-2",
      "date": "2023-03-20",
      "model": "Gen-2",
      "lab": "Runway",
      "lab_slug": "runway",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Multimodal video generation system that added pure text-to-video to Runway's Gen line, alongside text plus image to video, image to video and video-to-video stylization modes.",
      "highlights": [
        "Announced by Runway as Text to Video with Gen-2",
        "Modes include text to video, text plus image to video and image to video",
        "Also stylizes existing videos from an image or text prompt"
      ],
      "source_title": "Runway on X: Introducing, Text to Video. With Gen-2.",
      "source_url": "https://x.com/runwayml/status/1637800500459458562",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/03#runway-gen-2"
    },
    {
      "slug": "ernie-bot",
      "date": "2023-03-16",
      "model": "ERNIE Bot",
      "lab": "Baidu",
      "lab_slug": "baidu",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text",
        "image",
        "audio",
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "ERNIE Bot is Baidu's large language model and generative AI product, launched with abilities in literary creation, business writing, mathematical calculation, Chinese language understanding, and multimodal generation.",
      "highlights": [
        "Opened to invited users with invitation codes on March 16, 2023",
        "Enterprise API access offered on application through Baidu AI Cloud",
        "Baidu demonstrated generation of text, images, audio, and video from text prompts"
      ],
      "source_title": "Baidu Unveils ERNIE Bot, the Latest Generative AI Mastering Chinese Language and Multi-Modal Generation",
      "source_url": "https://www.prnewswire.com/news-releases/baidu-unveils-ernie-bot-the-latest-generative-ai-mastering-chinese-language-and-multi-modal-generation-301774240.html",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/03#ernie-bot"
    },
    {
      "slug": "midjourney-v5",
      "date": "2023-03-15",
      "model": "Midjourney V5",
      "lab": "Midjourney",
      "lab_slug": "midjourney",
      "type": "image",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "image"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Fifth-generation Midjourney image model with very high coherency, better interpretation of natural language prompts, higher resolution output and support for repeating patterns with --tile.",
      "highlights": [
        "Default model from March 30, 2023 to May 3, 2023",
        "Supported any aspect ratio, up from 2:1 or 1:2 on V4",
        "Produced more photographic images than V5.1 but often needed longer prompts"
      ],
      "source_title": "Legacy Features - Midjourney",
      "source_url": "https://docs.midjourney.com/hc/en-us/articles/33329788681101-Legacy-Features",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/03#midjourney-v5"
    },
    {
      "slug": "claude-1",
      "date": "2023-03-14",
      "model": "Claude and Claude Instant",
      "lab": "Anthropic",
      "lab_slug": "anthropic",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "Claude is Anthropic's first publicly offered AI assistant model, released after a closed alpha with partners such as Notion, Quora, and DuckDuckGo. It launched in two versions: Claude and the lighter, faster, less expensive Claude Instant.",
      "highlights": [
        "Accessible through a chat interface and the API in Anthropic's developer console",
        "Access by request at launch",
        "Offered through partner products including Quora's Poe, Notion AI, and DuckDuckGo's DuckAssist"
      ],
      "source_title": "Introducing Claude",
      "source_url": "https://www.anthropic.com/news/introducing-claude",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/03#claude-1"
    },
    {
      "slug": "gpt-4",
      "date": "2023-03-14",
      "model": "GPT-4",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 8192,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app",
        "API"
      ],
      "summary": "GPT-4 is a large multimodal model from OpenAI that accepts image and text inputs and emits text, released with text input in ChatGPT Plus and through a waitlisted API, with image input held back as a research preview.",
      "highlights": [
        "API model gpt-4 (snapshot gpt-4-0314) with an 8,192-token context",
        "gpt-4-32k with a 32,768-token context offered in limited access",
        "Developers can set the model's style and task through the system message",
        "API general availability for developers followed in July 2023"
      ],
      "source_title": "GPT-4",
      "source_url": "https://openai.com/index/gpt-4-research/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/03#gpt-4"
    },
    {
      "slug": "chatglm-6b",
      "date": "2023-03-13",
      "model": "ChatGLM-6B",
      "lab": "Z.ai",
      "lab_slug": "z-ai",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": 2000,
      "parameters": "6.2B",
      "model_license": "ChatGLM-6B License",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "ChatGLM-6B is Z.ai's open bilingual Chinese and English dialogue model with 6.2 billion parameters, based on the GLM architecture. It is designed for local deployment on consumer GPUs.",
      "highlights": [
        "6.2 billion parameters",
        "2K context length, later extended in ChatGLM2-6B",
        "Code under Apache 2.0; weights under the ChatGLM-6B License"
      ],
      "source_title": "ChatGLM-6B (GitHub)",
      "source_url": "https://github.com/zai-org/ChatGLM-6B",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/03#chatglm-6b"
    },
    {
      "slug": "jurassic-2",
      "date": "2023-03-09",
      "model": "Jurassic-2",
      "lab": "AI21 Labs",
      "lab_slug": "ai21",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "AI21's second generation of foundation models, offered as Large, Grande and Jumbo base models plus instruction-tuned Grande and Jumbo versions through its API and playground.",
      "highlights": [
        "Supports Spanish, French, German, Portuguese, Italian and Dutch",
        "Announced alongside AI21's task-specific APIs",
        "Available in the AI21 Studio playground and API"
      ],
      "source_title": "Announcing Jurassic-2 and Task-Specific APIs",
      "source_url": "https://www.ai21.com/blog/introducing-j2/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/03#jurassic-2"
    },
    {
      "slug": "gpt-3-5-turbo",
      "date": "2023-03-01",
      "model": "GPT-3.5 Turbo (gpt-3.5-turbo)",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "llm",
      "weights": "proprietary",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "gpt-3.5-turbo brought the model family behind ChatGPT to the API through a new Chat Completions endpoint that takes a sequence of messages, and OpenAI positioned it for many non-chat use cases as well.",
      "highlights": [
        "Same model family used in the ChatGPT product",
        "Pinned snapshot gpt-3.5-turbo-0301 released the same day",
        "Messages are rendered to the model in a format called Chat Markup Language (ChatML)",
        "Dedicated instances offered for customers who want reserved capacity"
      ],
      "source_title": "Introducing ChatGPT and Whisper APIs",
      "source_url": "https://openai.com/index/introducing-chatgpt-and-whisper-apis/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/03#gpt-3-5-turbo"
    },
    {
      "slug": "whisper-1",
      "date": "2023-03-01",
      "model": "Whisper API (whisper-1)",
      "lab": "OpenAI",
      "lab_slug": "openai",
      "type": "speech",
      "weights": "proprietary",
      "input_modalities": [
        "audio"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "API"
      ],
      "summary": "OpenAI made its open-source Whisper large-v2 speech recognition model available as a hosted API model, whisper-1, through transcription and translation endpoints backed by an optimized serving stack.",
      "highlights": [
        "Transcriptions endpoint keeps the source language; translations endpoint outputs English",
        "Accepts m4a, mp3, mp4, mpeg, mpga, wav, and webm files",
        "Serves Whisper large-v2; OpenAI open-sourced Whisper in September 2022",
        "Declared generally available in July 2023 alongside the GPT-4 API"
      ],
      "source_title": "Introducing ChatGPT and Whisper APIs",
      "source_url": "https://openai.com/index/introducing-chatgpt-and-whisper-apis/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/03#whisper-1"
    },
    {
      "slug": "llama",
      "date": "2023-02-24",
      "model": "LLaMA",
      "lab": "Meta",
      "lab_slug": "meta",
      "type": "llm",
      "weights": "open-weights",
      "input_modalities": [
        "text"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": "7B, 13B, 33B, 65B",
      "model_license": "Non-commercial bespoke license",
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "LLaMA is Meta's first family of foundation language models, released to researchers under a noncommercial license in four sizes from 7B to 65B parameters.",
      "highlights": [
        "LLaMA 65B and 33B trained on 1.4 trillion tokens; 7B on one trillion",
        "Access granted case by case to academic researchers and selected organizations"
      ],
      "source_title": "Introducing LLaMA: A foundational, 65-billion-parameter large language model",
      "source_url": "https://ai.meta.com/blog/large-language-model-llama-meta-ai/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/02#llama"
    },
    {
      "slug": "runway-gen-1",
      "date": "2023-02-06",
      "model": "Gen-1",
      "lab": "Runway",
      "lab_slug": "runway",
      "type": "video",
      "weights": "proprietary",
      "input_modalities": [
        "text",
        "image",
        "video"
      ],
      "output_modalities": [
        "video"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Consumer app"
      ],
      "summary": "Runway video-to-video model that applies the composition and style of an image or text prompt to the structure of a source video, generating new videos out of existing ones.",
      "highlights": [
        "Modes include Stylization, Storyboard, Mask, Render and Customization",
        "Early access opened on February 8, 2023",
        "Described in the paper Structure and Content-Guided Video Synthesis with Diffusion Models"
      ],
      "source_title": "Gen-1 | Runway Research",
      "source_url": "https://runway.com/research/gen-1",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/02#runway-gen-1"
    },
    {
      "slug": "blip-2",
      "date": "2023-01-30",
      "model": "BLIP-2",
      "lab": "Salesforce",
      "lab_slug": "salesforce",
      "type": "multimodal",
      "weights": "open-weights",
      "input_modalities": [
        "text",
        "image"
      ],
      "output_modalities": [
        "text"
      ],
      "context_window": null,
      "parameters": null,
      "model_license": null,
      "available_at_launch": [
        "Open weights"
      ],
      "summary": "Vision-language pre-training method and model family that connects frozen image encoders to frozen large language models through a lightweight Querying Transformer, enabling zero-shot instructed image-to-text generation.",
      "highlights": [
        "Turned an 11B LLM into a multimodal model by training only 188M parameters",
        "Released with OPT and FlanT5 language model variants",
        "Code and models were open-sourced in Salesforce's LAVIS library and integrated into Hugging Face Transformers"
      ],
      "source_title": "BLIP-2: Scalable Multimodal Pre-training Method",
      "source_url": "https://www.salesforce.com/blog/blip-2/",
      "retired_date": null,
      "retirement_note": null,
      "retirement_source_url": null,
      "permalink": "https://theaiengineering.com/releases/2023/01#blip-2"
    }
  ]
}