On October 29, 2026, Mistral AI published the open weights for Mistral Large 4 (“Le Chonk”), the model it opened as a public preview on October 6. Mistral had said the weights would arrive by the end of October.
What to know
- The model: Mistral describes it as a 1 trillion parameter, natively multimodal hybrid instruct and reasoning mixture-of-experts model, fluent in 160+ languages. The Hugging Face repo name points to about 52B active parameters. Mistral says it was developed and trained from scratch in Europe.
- Licence: the Mistral Ecosystem Open-Weight licence, a modified MIT licence. Mistral says you can use, modify, fine-tune and commercialise the model. Businesses that run a Model-as-a-Service offering need a separate commercial licence from Mistral.
- Reasoning modes: Low for simple tasks, Medium for everyday tasks and High, the default, for the hardest tasks.
- API pricing (Mistral’s figures): $0.70 per million input tokens, $0.07 per million cached input tokens and $2.10 per million output tokens.
- Where to get it: Mistral lists Mistral Studio, the Mistral API, OpenRouter, Microsoft Foundry, NVIDIA-hosted endpoints and Vibe Code for hosted use. For self-hosting it lists Hugging Face weights with vLLM or SGLang, plus an NVIDIA NIM container.
- Benchmarks: Mistral reports results such as 41% on Terminal-Bench 4.0 and 84.5% on BrowseComp. These are Mistral’s own figures, several using third-party evaluations, so wait for independent tests. Mistral says a full technical report is coming soon.
Why it matters
This turns Large 4 from a preview into something developers can download and run. Check the licence before you build a product on it, especially if you resell model access. New to the term? Read what open-weight models are, then see how to choose a Large 4 reasoning mode and hosting option. More on the Mistral page, the DeepSeek vs Mistral comparison and the model tracker.
Source: Mistral AI. We summarise in our own words; check the source for the full details.