Mistral releases 1T-parameter multimodal model with open weights
Mistral released Mistral Large 4, a 1 trillion-parameter multimodal model with 49 billion active parameters and a 1 million-token context window. The model supports text and image inputs and generates text, released under the Mistral Ecosystem Open-Weight License, an MIT derivative with usage restrictions. It excels in cybersecurity, coding agents, and multimodal tasks, scoring 82% on the…
Key points
- Mistral Large 4 is a 1T-parameter multimodal model with 49B active parameters, released under the Mistral Ecosystem Open-Weight License
- Scores 82% on Artificial Analysis Cyber Index, highest for reproducing and patching real software vulnerabilities, per Mistral’s internal benchmarks
- Supports text + image inputs, excels in coding (61.7% on DeepSWE v1.1) and agentic workflows, with GGUF quants and Hugging Face deployment options
Mistral Large 4 combines Mixture-of-Experts (MoE) architecture with sparse attention mechanisms, including a Pixtral vision encoder for image processing. The model’s text decoder has 72 layers, a hidden dimension of 6,144, and 64 attention heads, while the Pixtral encoder uses 14×14 patches and a patch-merge projector. Benchmarks show strong performance in coding (61.7% on DeepSWE v1.1) and agentic workflows (49.8% on Coding Agent Index), surpassing competitors like DeepSeek V4 Pro 0813 and Qwen3.8 Max. Mistral positions the model for defensive security work, noting that open weights enable custom deployment and fine-tuning without vendor lock-in. The model is available via Hugging Face, with GGUF quantizations provided by Unsloth and llama.cpp, and supports deployment on NVIDIA Blackwell GPUs via vLLM and DSpark for accelerated inference.
Model page: Mistral Large 4 →
The story so far
2 episodes →- Mistral releases 1T-parameter multimodal model with open weightsthis story
Welcome Le Chonk: Mistral Large 4 on Hugging Face
Hugging Face · 29 October 2026
Loading the full article…
This text was published by Hugging Face and written by merve. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
Coverage and discussion
7sources- Hacker News discussion · 115 pointsnews.ycombinator.com
- Reddit discussionreddit.com
- Reddit discussionreddit.com
- Mistral’s 1-Trillion-Parameter Le Chonk Checkpoint Lands on Hugging FacePress · Unite.AI ·
- Mistral releases open weights for its 1 trillion-parameter Large 4 modelPress · thenextweb.com ·
- Mistral releases the weights for Mistral Large 4, or Le Chonk, and says it rivals top open models such as GLM-5.3 on the Artificial Analysis Coding Agent IndexPress · zdnet.com ·
- mistralai/Mistral-Large-4-1T-A52B · Hugging FacePrimary source · huggingface.co ·
- Mistral Says Its AI Model ‘Le Chonk’ Has Leapfrogged China in the Open-Weight RacePress · Wired AI ·
- Le Chonk: Built for Coders, Trusted by Defenders.Primary source · Mistral AI ·
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- Moonshot AI and DeepSeek reportedly used GPT-6 Astra outputs to build new models · 6 src
- Gemini 4 Carbon tops Artificial Analysis Intelligence Index v5 · 2 src
- Google announces Gemini 4 Carbon and Genie 4 models with staged rollout for cyber defenders · 5 src
- Google launches Gemini 4 Carbon for cybersecurity defenders at $2/1M input and $10/1M output tokens · 10 src
- HemingwayAI model passes AI detectors 90% of the time in tests · 1 src
Comments
via GitHub Discussions