AI models released in November 2025
29 releases from 17 labs in November 2025, including DeepSeekMath-V2, Step1X-Edit v1.2 and FLUX.2 [dev]. Every entry is dated to the lab's official announcement and links to it.
- Releases
- 29
- Labs
- 17
- Open weights
- 17
November 2025
29 releasesDeepSeekMath-V2
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.
- Modalities
- text → text
- License
- Apache-2.0
- At launch
- Open weights
- 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
Step1X-Edit v1.2
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.
- Modalities
- text, image → image
- License
- Apache-2.0
- At launch
- Open weights
- 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
FLUX.2 [dev]
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.
- Modalities
- text, image → image
- Parameters
- 32B
- License
- FLUX Non-Commercial License
- At launch
- Open weights, Cloud platforms
- 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
FLUX.2 [pro] and [flex]
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.
- Modalities
- text, image → image
- At launch
- API, Consumer app, Cloud platforms
- 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
HunyuanOCR
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.
- Modalities
- text, image → text
- Parameters
- 1B
- License
- Tencent Hunyuan Community License Agreement
- At launch
- Open weights
- 1B parameters
- Supports over 100 languages, including mixed-language documents
Claude Opus 4.5
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.
- Modalities
- text, image → text
- Context
- 200K context
- At launch
- API, Consumer app, Cloud platforms
- 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
Fara-7B
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.
- Modalities
- text, image → text
- Context
- 125K context
- Parameters
- 7B
- License
- MIT
- At launch
- Open weights, API
- 7B parameters with a 128K context length
- Builds on Qwen 2.5 VL
- Available on Hugging Face and Microsoft Foundry
HunyuanVideo 1.5
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.
- Modalities
- text, image → video
- Parameters
- 8.3B
- License
- Tencent Hunyuan Community License Agreement
- At launch
- Open weights
- 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
Keye-VL-671B-A37B
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.
- Modalities
- text, image, video → text
- Parameters
- MoE: 671B total, 37B active
- License
- Apache-2.0
- At launch
- Open weights
- Reuses the Keye-VL-1.5 vision encoder
- Synthetic chain-of-thought data used during continual pre-training
- Weights released under Apache-2.0
Nano Banana Pro (Gemini 3 Pro Image)
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.
- Modalities
- text, image → text, image
- Context
- 64K context
- At launch
- API, Cloud platforms, Consumer app, Enterprise
- 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
Olmo 3
Third generation of fully open Olmo models, released as Base, Think, Instruct and RL Zero variants with checkpoints from every major stage of training.
- Modalities
- text → text
- Parameters
- 7B and 32B
- License
- Apache-2.0
- At launch
- Open weights
- Pretrained on the new Dolma 3 data
- Releases base, mid-trained, long-context and post-training checkpoints
- Released under Apache 2.0
GPT-5.1-Codex-Max
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.
- Modalities
- text, image → text
- Context
- 400K context
- At launch
- Consumer app, Enterprise
- 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
Grok 4.1 Fast
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.
- Modalities
- text → text
- Context
- 2M context
- At launch
- API
- 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
MiMo-Embodied
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.
- Modalities
- text, image, video → text
- Parameters
- 7B
- License
- MIT
- At launch
- Open weights
- Released as MiMo-Embodied-7B on Hugging Face
- Evaluated on 17 embodied AI and 12 autonomous driving benchmarks
- Code and evaluation configs on GitHub
SAM 3
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.
- Modalities
- text, image, video → image, video
- Parameters
- 848M
- License
- SAM License
- At launch
- Open weights
- 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
SAM 3D (Objects and Body)
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.
- Modalities
- image → 3d
- License
- SAM License
- At launch
- Open weights
- 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
Step-Audio-R1
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.
- Modalities
- audio, text → text
- License
- Apache-2.0
- At launch
- Open weights
- 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
Gemini 3 Pro
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.
- Modalities
- text, image, audio, video → text
- Context
- 1M context
- At launch
- API, Cloud platforms, Consumer app
- 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
Grok 4.1
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.
- Modalities
- text → text
- At launch
- Consumer app
- 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
Depth Anything 3
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.
- Modalities
- image, video → 3d
- Parameters
- 0.08B to 1.40B (model series)
- License
- Apache-2.0 (Small, Base, Metric, Mono); CC-BY-NC-4.0 (Large, Giant, Nested)
- At launch
- Open weights
- 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
ERNIE 5.0
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.
- Modalities
- text, image, audio, video → text, image, audio, video
- Parameters
- MoE: 2.4T total, about 3% active per token
- At launch
- Consumer app, API
- 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
GPT-5.1-Codex and GPT-5.1-Codex-Mini
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.
- Modalities
- text, image → text
- Context
- 400K context
- At launch
- API, Consumer app
- 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
GPT-5.1 (Instant and Thinking)
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.
- Modalities
- text, image → text
- At launch
- Consumer app, API, Enterprise
- 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
ERNIE-4.5-VL-28B-A3B-Thinking
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.
- Modalities
- text, image, video → text
- Parameters
- MoE: 28B total, 3B active
- License
- Apache-2.0
- At launch
- Open weights
- 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
Scribe v2 Realtime
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.
- Modalities
- audio → text
- At launch
- API
- Under 150ms latency
- Automatic language detection with mid-conversation switching
- Integrated into ElevenLabs Agents
Omnilingual ASR
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.
- Modalities
- audio → text
- Parameters
- 300M to 7B
- License
- Apache-2.0
- At launch
- Open weights
- 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
GPT-5-Codex-Mini
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.
- Modalities
- text → text
- At launch
- Consumer app
- 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
Kimi K2 Thinking
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.
- Modalities
- text → text
- Context
- 250K context
- Parameters
- MoE: 1T total, 32B active
- License
- Modified MIT License
- At launch
- API, Consumer app, Open weights
- 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
Step-Audio-EditX
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.
- Modalities
- audio, text → audio
- Parameters
- 3B
- License
- Apache-2.0
- At launch
- Open weights
- 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
Cite and reuse this data
These release records are free to reuse under CC BY 4.0, including commercially. Credit The AI Engineering and link to this page's URL.
Dates are the labs' official announcement dates, and every record links to its source. List fields in the CSV are separated by " | ".
Citation
The AI Engineering Editors. "AI models released in November 2025." The AI Engineering, updated Nov 27, 2025. https://theaiengineering.com/releases/2025/11
HTML link
<a href="https://theaiengineering.com/releases/2025/11">AI models released in November 2025</a> (The AI Engineering)- Full dataset
- 1,167 releases from 55 labs
- Coverage
- to
- Updated
- License
- CC BY 4.0
- Publisher
- The AI Engineering