Mistral Large 4 Preview: Canary-Test Silent Model Updates

Mistral shipped Large 4 (nicknamed Le Chonk) on October 6, 2026 as a public API preview. It is a 1.05 trillion parameter mixture-of-experts model with 49 billion active parameters, a 1.6B vision encoder, function calling, structured outputs, and an intro price of $0.68 per million input tokens and $2.09 per million output tokens (half the list price, with no published end date). Open weights are promised for the end of October. That combination is why people are trying it today.

Buried in the launch notes is the part that should change how you integrate it: Mistral says a preview model may receive silent updates without notice, carries no general-availability guarantee, and gets one month of deprecation notice. The model behind mistral-large-4 can be different on Friday than it was on Tuesday, and your extraction prompt, your tool schema and your cost model all quietly depend on it.