Mistral Large 4 launches on API, open weights due end of October
- Source
- x.com
- Date
Meet Mistral Large 4, aka Le Chonk. • 1T parameters, natively multimodal. 49B active. It is the best open weights model from US or Europe on aggregated benchmarks. • State-of-the-art on critical workloads, including cyber defense, manufacturing and finance and it surpasses closed frontier models on visual grounding. • Forged in Europe end-to-end and is deployable from Europe via our own Mistral Cloud infrastructure. • Available to all via API today. Working with cybersecurity partners privately. Open weights release end of October.
Context
Mistral AI released Mistral Large 4 on October 6 as a model anyone can call through its API, with the downloadable weights promised for the end of October. Its announcement on X describes a model with 1 trillion parameters, the adjustable numbers a model learns, of which only a fraction is active for any one input, and says it handles images natively. The post gives 49 billion as that active share, while Mistral's launch article says 52 billion and calls the release a public preview; the two first-party sources disagree on that figure.
The performance claims in the post are Mistral's own: the best open-weights model from the US or Europe on aggregated benchmarks, state of the art on cyber defense, manufacturing and finance, and ahead of closed frontier models on visual , meaning locating things precisely in an image. The post names no benchmarks. The launch article supports the visual-grounding claim with one result, 42 percent on Dense 200 against 41 percent for GPT-6 Astra, a one-point margin, but it does not name the aggregated behind the open-weights ranking.
An outside measurement gives a narrower picture. Artificial Analysis scored the preview 38 on its Intelligence Index, and its article places that alongside GPT-6 Luna at 38 and DeepSeek V4.1 Flash at 39, while reporting a cost per task more than four times that of open-weights models of similar intelligence. Only the opening post of Mistral's thread was collected and its video was not reviewed, so any benchmark detail shown there is not covered here.
- grounding — Tying a model's answers to checkable sources — retrieved documents, live data, tool results — instead of letting it answer from memory alone.
- benchmark — A standard public test set for comparing AI models — the shared scoreboards behind every "model X beats model Y" claim.
- multimodal — A model that works with more than text — reading images, audio, or video, and sometimes generating them too.
- open weights — A model whose trained parameters are published for anyone to download and run — unlike API-only models you can access but never possess.
Checking sign-in…
Loading comments…






