Mistral AI has released the weights of Mistral Large 4, the model it previewed in early October, so organisations can now download it from Hugging Face and run it on their own servers. Nicknamed “Le Chonk”, it is a mixture-of-experts model with about 1 trillion parameters, of which 52 billion are active for each task. It reads text and images and writes text and code.
The Paris company says further training since the preview improved its coding and cyber defence results; by its own figures, its Terminal-Bench 4.0 score rose 13 points to 41%. Mistral calls it “the best open-weight model for coding”, a vendor claim that has not been independently checked. The weights come under the Mistral Ecosystem Open-Weight licence, a modified MIT licence that allows commercial use and fine-tuning but requires a separate commercial licence for companies selling the model as a hosted service. Mistral lists API prices of $0.70 per million input tokens and $2.10 per million output tokens.
Why it matters: it gives organisations a very large model they can run and adapt on their own hardware. Mistral also stresses that it was trained from scratch on its own infrastructure in Europe, a selling point for governments and companies wary of depending on US or Chinese providers. Read Mistral’s announcement.