---
title: "AI models released in March 2026"
type: "release-archive"
canonical_url: "https://theaiengineering.com/releases/2026/03"
markdown_url: "https://theaiengineering.com/releases/2026/03.md"
author: "The AI Engineering Editors"
publisher: "The AI Engineering"
updated: "2026-03-31"
citation: "The AI Engineering Editors. \"AI models released in March 2026.\" The AI Engineering, updated Mar 31, 2026. https://theaiengineering.com/releases/2026/03"
license: "CC BY 4.0"
license_url: "https://creativecommons.org/licenses/by/4.0/"
data_csv: "https://theaiengineering.com/releases.csv"
data_json: "https://theaiengineering.com/releases.json"
---

# AI models released in March 2026

42 releases from 22 labs, 21 with open weights. Dates are the labs' official announcement dates.

## March 2026

### LFM2.5-350M

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.

- Date: 2026-03-31
- Lab: Liquid AI
- Type: Small model
- Weights: Open weights
- Modalities: text → text
- Context window: 32,768 tokens
- Parameters: 350M
- License: LFM Open License v1.0
- Available at launch: Consumer app, Open weights
- Permalink: https://theaiengineering.com/releases/2026/03#lfm2-5-350m

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: [LFM2.5-350M: No Size Left Behind](https://www.liquid.ai/blog/lfm2-5-350m-no-size-left-behind)

### Veo 3.1 Lite

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.

- Date: 2026-03-31
- Lab: Google
- Type: Video
- Weights: Proprietary
- Modalities: text, image → video
- Available at launch: API, Cloud platforms
- Permalink: https://theaiengineering.com/releases/2026/03#veo-3-1-lite

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: [Build with Veo 3.1 Lite, our most cost-effective video generation model](https://blog.google/innovation-and-ai/technology/ai/veo-3-1-lite/)

### LongCat-AudioDiT

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.

- Date: 2026-03-30
- Lab: Meituan
- Type: Speech
- Weights: Open weights
- Modalities: text, audio → audio
- License: MIT
- Available at launch: Open weights
- Permalink: https://theaiengineering.com/releases/2026/03#longcat-audiodit

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: [GitHub - meituan-longcat/LongCat-AudioDiT](https://github.com/meituan-longcat/LongCat-AudioDiT)

### Qwen3.5-Omni

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.

- Date: 2026-03-30
- Lab: Alibaba Qwen
- Type: Multimodal
- Weights: Proprietary
- Modalities: text, image, audio, video → text, audio
- Context window: 256,000 tokens
- Available at launch: API
- Permalink: https://theaiengineering.com/releases/2026/03#qwen3-5-omni

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: [Qwen3.5-Omni: Scaling Up, Toward Native Omni-Modal AGI](https://qwen.ai/blog?id=qwen3.5-omni)

### Granite 4.0 3B Vision

Vision-language model from IBM designed for enterprise-grade document data extraction, focused on specialized, complex extraction tasks that ultracompact models often struggle with.

- Date: 2026-03-27
- Lab: IBM
- Type: Multimodal
- Weights: Open weights
- Modalities: text, image → text
- License: Apache-2.0
- Available at launch: Open weights
- Permalink: https://theaiengineering.com/releases/2026/03#granite-4-0-3b-vision

Highlights:

- Focused on document data extraction
- Predecessor to Granite Vision 4.1
- Released under Apache 2.0

Source: [ibm-granite/granite-4.0-3b-vision · Hugging Face](https://huggingface.co/ibm-granite/granite-4.0-3b-vision)

### SAM 3.1

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.

- Date: 2026-03-27
- Lab: Meta
- Type: Multimodal
- Weights: Open weights
- Modalities: text, image, video → image, video
- License: SAM License
- Available at launch: Open weights, API
- Permalink: https://theaiengineering.com/releases/2026/03#meta-sam-3-1

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: [SAM 3.1: Faster and More Accessible Real-Time Video Detection and Tracking With Multiplexing and Global Reasoning](https://ai.meta.com/blog/segment-anything-model-3/)

### Cohere Transcribe

Cohere's first transcription model, an automatic speech recognition model that converts audio to text in 14 languages through the Audio Transcriptions API.

- Date: 2026-03-26
- Lab: Cohere
- Type: Speech
- Weights: Open weights
- Modalities: audio → text
- License: Apache-2.0
- Available at launch: API, Open weights
- Permalink: https://theaiengineering.com/releases/2026/03#cohere-transcribe

Highlights:

- Model name cohere-transcribe-03-2026
- Languages include English, German, French, Japanese, Korean, Chinese and Arabic
- Released under Apache 2.0

Source: [Announcing the Cohere Transcribe model](https://docs.cohere.com/v2/changelog/cohere-transcribe-03-2026)

### Gemini 3.1 Flash Live

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.

- Date: 2026-03-26
- Lab: Google
- Type: Speech
- Weights: Proprietary
- Modalities: text, image, audio → text, audio
- Context window: 131,072 tokens
- Available at launch: API, Consumer app
- Permalink: https://theaiengineering.com/releases/2026/03#gemini-3-1-flash-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: [Gemini 3.1 Flash Live: Google's latest AI audio model](https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-flash-live/)

### Suno v5.5

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.

- Date: 2026-03-26
- Lab: Suno
- Type: Music
- Weights: Proprietary
- Modalities: text, audio → audio
- Available at launch: Consumer app
- Permalink: https://theaiengineering.com/releases/2026/03#suno-v5-5

Highlights:

- Voices available to Pro and Premier subscribers
- Custom Models for a personalized sound
- My Taste personalization

Source: [Suno v5.5: More Expressive. More You.](https://suno.com/blog/v5-5)

### LongCat-Next

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.

- Date: 2026-03-25
- Lab: Meituan
- Type: Multimodal
- Weights: Open weights
- Modalities: text, image, audio → text, image, audio
- License: MIT
- Available at launch: Open weights
- Permalink: https://theaiengineering.com/releases/2026/03#longcat-next

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: [GitHub - meituan-longcat/LongCat-Next](https://github.com/meituan-longcat/LongCat-Next)

### Lyria 3 Pro

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.

- Date: 2026-03-25
- Lab: Google
- Type: Music
- Weights: Proprietary
- Modalities: text → audio
- Available at launch: Consumer app, API, Cloud platforms
- Permalink: https://theaiengineering.com/releases/2026/03#lyria-3-pro

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: [Lyria 3 Pro: Create longer tracks in more Google products](https://blog.google/innovation-and-ai/technology/ai/lyria-3-pro/)

### MolmoWeb

Open visual web agent built on Molmo 2 that navigates websites, fills out forms and retrieves information in a browser, designed for self-hosting.

- Date: 2026-03-24
- Lab: Ai2
- Type: Multimodal
- Weights: Open weights
- Modalities: text, image → text
- Parameters: 4B and 8B
- License: Apache-2.0
- Available at launch: Open weights
- Permalink: https://theaiengineering.com/releases/2026/03#molmoweb

Highlights:

- Released with weights, training data and evaluation tools
- Training data includes 36K human task trajectories
- Released under Apache 2.0

Source: [MolmoWeb: An open agent for automating web tasks | Ai2](https://allenai.org/blog/molmoweb)

### Voxtral TTS

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.

- Date: 2026-03-23
- Lab: Mistral AI
- Type: Speech
- Weights: Open weights
- Modalities: text, audio → audio
- Parameters: 4B
- License: CC-BY-NC-4.0
- Available at launch: Open weights, API
- Permalink: https://theaiengineering.com/releases/2026/03#voxtral-tts

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: [Speaking of Voxtral | Mistral AI](https://mistral.ai/news/voxtral-tts)

### LongCat-Flash-Prover

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.

- Date: 2026-03-20
- Lab: Meituan
- Type: Reasoning
- Weights: Open weights
- Modalities: text → text
- Parameters: MoE: 560B total
- License: MIT
- Available at launch: Open weights
- Permalink: https://theaiengineering.com/releases/2026/03#longcat-flash-prover

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: [GitHub - meituan-longcat/LongCat-Flash-Prover](https://github.com/meituan-longcat/LongCat-Flash-Prover)

### MAI-Image-2

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.

- Date: 2026-03-19
- Lab: Microsoft
- Type: Image
- Weights: Proprietary
- Modalities: text → image
- Available at launch: API, Consumer app
- Permalink: https://theaiengineering.com/releases/2026/03#mai-image-2

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: [Introducing MAI-Image-2: for limitless creativity](https://microsoft.ai/news/introducing-mai-image-2/)

### MiMo-V2-Omni

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.

- Date: 2026-03-18
- Lab: Xiaomi
- Type: Multimodal
- Weights: Proprietary
- Modalities: text, image, video, audio → text
- Context window: 256,000 tokens
- Available at launch: API, Consumer app
- Permalink: https://theaiengineering.com/releases/2026/03#mimo-v2-omni

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: [MiMo-V2-Omni | Xiaomi](https://mimo.xiaomi.com/mimo-v2-omni/index.html)

### MiMo-V2-Pro

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.

- Date: 2026-03-18
- Lab: Xiaomi
- Type: LLM
- Weights: Proprietary
- Modalities: text → text
- Context window: 1,000,000 tokens
- Parameters: MoE: over 1T total, 42B active
- Available at launch: API, Consumer app
- Permalink: https://theaiengineering.com/releases/2026/03#mimo-v2-pro

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: [MiMo-V2-Pro | Xiaomi](https://mimo.xiaomi.com/mimo-v2-pro/index.html)

### MiMo-V2-TTS

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.

- Date: 2026-03-18
- Lab: Xiaomi
- Type: Speech
- Weights: Proprietary
- Modalities: text → audio
- Available at launch: API, Consumer app
- Permalink: https://theaiengineering.com/releases/2026/03#mimo-v2-tts

Highlights:

- Supports style control, singing, and voice cloning
- Refined with multi-dimensional reinforcement learning
- Launch demo covers Northeastern Mandarin, Sichuanese, and Cantonese speech

Source: [MiMo-V2-TTS | Xiaomi](https://mimo.xiaomi.com/mimo-v2-tts/index.html)

### MiniMax M2.7

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.

- Date: 2026-03-18
- Lab: MiniMax
- Type: LLM
- Weights: Open weights
- Modalities: text → text
- License: Non-Commercial License
- Available at launch: API, Open weights
- Permalink: https://theaiengineering.com/releases/2026/03#minimax-m2-7

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: [MiniMax M2.7: Early Echoes of Self-Evolution](https://www.minimax.io/news/minimax-m27-en)

### Qianfan-OCR

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.

- Date: 2026-03-18
- Lab: Baidu
- Type: Multimodal
- Weights: Open weights
- Modalities: image, text → text
- Parameters: 4B
- License: Apache-2.0
- Available at launch: Open weights
- Permalink: https://theaiengineering.com/releases/2026/03#qianfan-ocr

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: [GitHub - baidubce/Qianfan-VL: Qianfan-VL: Domain-Enhanced Universal Vision-Language Models](https://github.com/baidubce/Qianfan-VL)

### GPT-5.4 mini and GPT-5.4 nano

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.

- Date: 2026-03-17
- Lab: OpenAI
- Type: Small model
- Weights: Proprietary
- Modalities: text, image → text
- Context window: 400,000 tokens
- Available at launch: API, Consumer app
- Permalink: https://theaiengineering.com/releases/2026/03#gpt-5-4-mini

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: [Introducing GPT-5.4 mini and nano | OpenAI](https://openai.com/index/introducing-gpt-5-4-mini-and-nano/)

### Midjourney V8 (Alpha)

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.

- Date: 2026-03-17
- Lab: Midjourney
- Type: Image
- Weights: Proprietary
- Modalities: text, image → image
- Available at launch: Consumer app
- Permalink: https://theaiengineering.com/releases/2026/03#midjourney-v8-alpha

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: [V8 Alpha](https://updates.midjourney.com/v8-alpha/)

### Leanstral

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.

- Date: 2026-03-16
- Lab: Mistral AI
- Type: Coding
- Weights: Open weights
- Modalities: text → text
- Context window: 256,000 tokens
- Parameters: MoE: 119B total, 6.5B active
- License: Apache-2.0
- Available at launch: Open weights, API
- Permalink: https://theaiengineering.com/releases/2026/03#leanstral

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: [Leanstral: Open-Source foundation for trustworthy vibe-coding | Mistral AI](https://mistral.ai/news/leanstral)

### Mistral Small 4

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.

- Date: 2026-03-16
- Lab: Mistral AI
- Type: Multimodal
- Weights: Open weights
- Modalities: text, image → text
- Context window: 256,000 tokens
- Parameters: MoE: 119B total, 6B active
- License: Apache-2.0
- Available at launch: Open weights, API
- Permalink: https://theaiengineering.com/releases/2026/03#mistral-small-4

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: [Introducing Mistral Small 4 | Mistral AI](https://mistral.ai/news/mistral-small-4)

### Nemotron 3 Nano 4B

Compact member of the Nemotron 3 family with about 4B parameters and a long context window, for local and edge deployments.

- Date: 2026-03-16
- Lab: NVIDIA
- Type: Small model
- Weights: Open weights
- Modalities: text → text
- Context window: 262,000 tokens
- Parameters: 3.97B
- License: NVIDIA Nemotron Open Model License
- Available at launch: Open weights
- Permalink: https://theaiengineering.com/releases/2026/03#nemotron-3-nano-4b

Highlights:

- Context length up to 262K tokens
- Released under the NVIDIA Nemotron Open Model License
- Smaller sibling of Nemotron 3 Nano 30B

Source: [nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16 · Hugging Face](https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16)

### GLM-5-Turbo

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.

- Date: 2026-03-15
- Lab: Z.ai
- Type: LLM
- Weights: Proprietary
- Modalities: text → text
- Context window: 200,000 tokens
- Available at launch: API
- Permalink: https://theaiengineering.com/releases/2026/03#glm-5-turbo

Highlights:

- Improved decomposition of complex instructions and multi-agent coordination
- Up to 128K output tokens
- Available through the BigModel and Z.ai APIs

Source: [Model and product release records (BigModel docs)](https://docs.bigmodel.cn/cn/update/new-releases)

### Mistral Moderation 2

Second-generation Mistral moderation model with a 128K-token context window and jailbreak detection, built for complex multilingual content including long multi-turn conversations.

- Date: 2026-03-12
- Lab: Mistral AI
- Type: LLM
- Weights: Proprietary
- Modalities: text → text
- Context window: 128,000 tokens
- Available at launch: API
- Permalink: https://theaiengineering.com/releases/2026/03#mistral-moderation-2

Highlights:

- API model ID mistral-moderation-2603
- Released with custom guardrails for Agents and Conversations
- Mistral's model card lists March 1, 2026

Source: [Changelog | Mistral Docs](https://docs.mistral.ai/resources/changelogs)

### LongCat-Flash-Omni-2603

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.

- Date: 2026-03-11
- Lab: Meituan
- Type: Multimodal
- Weights: Proprietary
- Modalities: text, image, audio, video → text, audio
- Available at launch: API, Consumer app
- Permalink: https://theaiengineering.com/releases/2026/03#longcat-flash-omni-2603

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: [Change Log | LongCat API Docs](https://longcat.chat/platform/docs/change-log)

### Nemotron 3 Super

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.

- Date: 2026-03-11
- Lab: NVIDIA
- Type: Reasoning
- Weights: Open weights
- Modalities: text → text
- Context window: 1,000,000 tokens
- Parameters: MoE: 120B total, 12B active
- License: NVIDIA Nemotron Open Model License
- Available at launch: Open weights
- Permalink: https://theaiengineering.com/releases/2026/03#nemotron-3-super

Highlights:

- Adds multi-token prediction layers
- Supports up to 1M tokens of context
- Trained between December 2025 and March 2026

Source: [nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16 · Hugging Face](https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16)

### Reka Edge (reka-edge-2603)

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.

- Date: 2026-03-11
- Lab: Reka
- Type: Small model
- Weights: Open weights
- Modalities: text, image, video → text
- Context window: 16,384 tokens
- Parameters: 7B (657M vision encoder, 6.4B transformer)
- License: Business Source License 1.1 (Reka Edge 2603 License)
- Available at launch: API, Consumer app, Open weights
- Permalink: https://theaiengineering.com/releases/2026/03#reka-edge-2603

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: [Reka Edge: Frontier-Level Edge Intelligence for Physical AI](https://reka.ai/news/reka-edge-frontier-level-edge-intelligence-for-physical-ai)

### Gemini Embedding 2

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.

- Date: 2026-03-10
- Lab: Google
- Type: Embedding
- Weights: Proprietary
- Modalities: text, image, video, audio → embedding
- Context window: 8,192 tokens
- Available at launch: API, Cloud platforms
- Permalink: https://theaiengineering.com/releases/2026/03#gemini-embedding-2

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: [Gemini Embedding 2: Our first natively multimodal embedding model](https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-embedding-2/)

### Grok 4.20 and Grok 4.20 Multi-agent

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.

- Date: 2026-03-10
- Lab: xAI
- Type: LLM
- Weights: Proprietary
- Modalities: text, image → text
- Context window: 1,000,000 tokens
- Available at launch: API
- Permalink: https://theaiengineering.com/releases/2026/03#grok-4-20

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: [Release Notes | SpaceXSI Docs](https://docs.x.ai/developers/release-notes)

### Runway Characters (GWM-1 Avatars)

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.

- Date: 2026-03-09
- Lab: Runway
- Type: Video
- Weights: Proprietary
- Modalities: image, audio, text → video, audio
- Available at launch: API, Enterprise
- Permalink: https://theaiengineering.com/releases/2026/03#runway-characters

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: [Introducing Runway Characters](https://runway.com/news/introducing-runway-characters)

### Granite 4.0 1B Speech

Compact speech-language model for multilingual automatic speech recognition and bidirectional speech translation, with half the parameters of Granite Speech 3.3 2B.

- Date: 2026-03-06
- Lab: IBM
- Type: Speech
- Weights: Open weights
- Modalities: audio → text
- License: Apache-2.0
- Available at launch: Open weights
- Permalink: https://theaiengineering.com/releases/2026/03#granite-4-0-1b-speech

Highlights:

- Built for resource-constrained devices
- Supports Japanese and keyword-biased recognition
- Released under Apache 2.0

Source: [ibm-granite/granite-4.0-1b-speech · Hugging Face](https://huggingface.co/ibm-granite/granite-4.0-1b-speech)

### GPT-5.4 (Thinking and Pro)

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.

- Date: 2026-03-05
- Lab: OpenAI
- Type: LLM
- Weights: Proprietary
- Modalities: text, image → text
- Context window: 1,050,000 tokens
- Available at launch: Consumer app, API, Enterprise
- Permalink: https://theaiengineering.com/releases/2026/03#gpt-5-4

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: [Introducing GPT-5.4 | OpenAI](https://openai.com/index/introducing-gpt-5-4/)

### LTX-2.3

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.

- Date: 2026-03-05
- Lab: Lightricks
- Type: Video
- Weights: Open weights
- Modalities: text, image, video, audio → video, audio
- Parameters: 22B
- License: LTX-2 Community License
- Available at launch: Open weights, API, Consumer app
- Permalink: https://theaiengineering.com/releases/2026/03#ltx-2-3

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: [Release Notes | LTX](https://ltx.io/release-notes)

### Olmo Hybrid

Fully open 7B model family that mixes transformer attention with linear recurrent layers, released with a controlled comparison against Olmo 3 7B.

- Date: 2026-03-05
- Lab: Ai2
- Type: LLM
- Weights: Open weights
- Modalities: text → text
- Parameters: 7B
- License: Apache-2.0
- Available at launch: Open weights
- Permalink: https://theaiengineering.com/releases/2026/03#olmo-hybrid

Highlights:

- Combines transformers and linear RNNs
- Developed through 1B and 7B scale experiments
- Released under Apache 2.0

Source: [Introducing Olmo Hybrid: Combining transformers and linear RNNs for superior scaling | Ai2](https://allenai.org/blog/olmohybrid)

### MiniMax Music 2.5+

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.

- Date: 2026-03-04
- Lab: MiniMax
- Type: Music
- Weights: Proprietary
- Modalities: text → audio
- Available at launch: API, Consumer app
- Permalink: https://theaiengineering.com/releases/2026/03#minimax-music-2-5-plus

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: [Music 2.5+: Unlock instrumental music, break through style boundaries](https://www.minimax.io/news/music-25-unlock-instrumental-music)

### Phi-4-reasoning-vision-15B

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.

- Date: 2026-03-04
- Lab: Microsoft
- Type: Multimodal
- Weights: Open weights
- Modalities: text, image → text
- Context window: 16,384 tokens
- Parameters: 15B
- License: MIT
- Available at launch: Open weights, API
- Permalink: https://theaiengineering.com/releases/2026/03#phi-4-reasoning-vision-15b

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: [Phi-4-reasoning-vision and the lessons of training a multimodal reasoning model](https://www.microsoft.com/en-us/research/blog/phi-4-reasoning-vision-and-the-lessons-of-training-a-multimodal-reasoning-model/)

### Gemini 3.1 Flash-Lite

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.

- Date: 2026-03-03
- Lab: Google
- Type: LLM
- Weights: Proprietary
- Modalities: text, image, audio, video → text
- Context window: 1,048,576 tokens
- Available at launch: API, Cloud platforms
- Permalink: https://theaiengineering.com/releases/2026/03#gemini-3-1-flash-lite

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: [Gemini 3.1 Flash Lite: Our most cost-effective AI model yet](https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-1-flash-lite/)

### GPT-5.3 Instant

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.

- Date: 2026-03-03
- Lab: OpenAI
- Type: LLM
- Weights: Proprietary
- Modalities: text, image → text
- Context window: 128,000 tokens
- Available at launch: Consumer app, API
- Permalink: https://theaiengineering.com/releases/2026/03#gpt-5-3-instant

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: [GPT-5.3 Instant: Smoother, more useful everyday conversations | OpenAI](https://openai.com/index/gpt-5-3-instant/)

### Qwen3.5 small models

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.

- Date: 2026-03-02
- Lab: Alibaba Qwen
- Type: Small model
- Weights: Open weights
- Modalities: text, image → text
- Context window: 262,144 tokens
- Parameters: 0.8B, 2B, 4B, 9B
- License: Apache-2.0
- Available at launch: Open weights
- Permalink: https://theaiengineering.com/releases/2026/03#qwen3-5-small

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: [Qwen3.8 (GitHub), news entry for Qwen3.5 small models](https://github.com/QwenLM/Qwen3.8)

## 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 " | ".

The AI Engineering Editors. "AI models released in March 2026." The AI Engineering, updated Mar 31, 2026. https://theaiengineering.com/releases/2026/03

HTML link:

```html
<a href="https://theaiengineering.com/releases/2026/03">AI models released in March 2026</a> (The AI Engineering)
```

- Title: AI models released in March 2026
- Author: The AI Engineering Editors
- Publisher: The AI Engineering
- Updated: 2026-03-31
- URL: https://theaiengineering.com/releases/2026/03
- Markdown: https://theaiengineering.com/releases/2026/03.md
- License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)
- Full dataset as CSV: https://theaiengineering.com/releases.csv
- Full dataset as JSON: https://theaiengineering.com/releases.json
