Mistral Releases Open Weights for Le Chonk, Its One Trillion Parameter Mistral Large 4 Model
Today, we release the weights of a new and improved Le Chonk, advancing the open frontier for agentic coding and cybersecurity. 💻 Built to build: Le Chonk ranks #1 on @cognition’s FrontierCode among open models, and sets a new benchmark for Pareto efficiency on @artificialanalysis. 🛡 Built to defend: State-of-the-art in cyberdefense, put through the paces by the world's leading cyber vendors, including @CrowdStrike, @cloudflare, @rubrik, @zscaler, and others. Built for sovereignty: Make it yours. Download the weights on @huggingface , customize it with Mistral Forge, and deploy it wherever you need it. Read more from our team and customers 👇:
Context
A European lab putting a trillion parameter model with a 1M token context into open weights gives companies and governments a strong coding and security model they can run on their own hardware.
Le Chonk is the nickname for Mistral Large 4, Mistral's largest model, which it says was trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in its own European datacenters. Open weights means anyone can download and run the model. The license is a modified MIT license that allows commercial use but requires a separate deal for companies that resell it as a hosted model service.
Deep dive
Mistral said on Thursday it is releasing the weights of a new and improved Le Chonk, the nickname for Mistral Large 4, pitching it at agentic coding and cybersecurity.
According to the Hugging Face model card, it is a natively multimodal mixture of experts model with about one trillion total parameters and 49 billion active per token, 128 routed experts with four active at a time, and a one million token context window. It takes text and images as input, offers low, medium and high reasoning effort, and ships in FP8 and NVFP4 versions.
Mistral claims Le Chonk ranks first among open models on Cognition's FrontierCode and sets a new mark for Pareto efficiency on Artificial Analysis. It also says cyber vendors including CrowdStrike, Cloudflare, Rubrik and Zscaler tested its defensive security skills. These are Mistral's own claims, and a full technical report is still to come.
The weights are on Hugging Face under the Mistral Ecosystem Open Weight License, which allows use, fine tuning and commercial products, but businesses that offer the model as a hosted service need a separate commercial license. On the Mistral API it costs 0.70 dollars per million input tokens and 2.10 dollars per million output tokens.
What to expect next
Watch for independent results on Artificial Analysis and coding leaderboards, and for the promised technical report. Running a trillion parameter model takes serious hardware, so adoption will depend on how well quantized versions and GGUF builds perform. The license limit on hosted services will also shape whether cloud providers offer it. European governments and regulated firms looking for sovereign AI are the most likely early users.