One-sentence summary
Mistral Large 4 (Le Chonk) opened weights on October 29, 2026: about 1.05T total params, 1M-token context, under a modified MIT license, which Mistral calls one of the strongest western open-weight models.
Basic Questions
1. When did Mistral Large 4 open weights?
October 29, 2026. Previewed on October 6, the weights were released on Hugging Face and served via the Mistral API, OpenRouter, NVIDIA NIM and Microsoft Foundry.
2. What is its parameter scale and architecture?
About 1.05T total params, ~52B active per token. A granular MoE with 128 routed experts, 4 active per token, plus 1 shared expert; natively multimodal with text and image input and text output.
3. How large is the context window?
1 million tokens. Trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in Mistral's own European datacenters, natively fluent in 160+ languages.
4. What is the weight license and commercial limit?
Mistral Ecosystem Open-Weight License (modified MIT). Free to use, modify, fine-tune and commercialize, but MaaS operators need a separate commercial license from Mistral.
Advanced Questions
5. How is the API priced?
Input $0.70/M tokens, cached input $0.07/M, output $2.10/M. Endpoint is mistral-large-4; an NVFP4 quantized checkpoint targets Blackwell GPUs.
6. Is it the strongest open-weight model?
It leads the western camp but is not the overall leader. Its Intelligence Index v5 score of 26 is the highest among western open weights, but the company concedes Chinese open-weight models (GLM, DeepSeek, Kimi) score higher on the aggregate index.
7. What resources does self-hosting need?
A sizable GPU cluster. At about 1T parameters, Mistral offers an NVFP4 quantized checkpoint to lower the bar; exact memory needs should be checked against the model card and target precision.
Sources
- Unite.AI: Mistral Le Chonk
- IT Home: Mistral open-sources Large 4 weights