Mistral releases the weights of Large 4, a 1 trillion parameter model

What did Mistral release?

Mistral released the weights of Mistral Large 4, nicknamed Le Chonk. Weights are the files that let anyone download a model and run it on their own hardware. The model first launched as a public preview on 6 October 2026, when Mistral promised the weights by the end of the month.

Mistral says the version it released today is improved, after three more weeks of reinforcement learning, and that it uses fewer tokens to do the same work.

How big is Mistral Large 4?

According to its model card, Mistral Large 4 has 1 trillion parameters, with 49 billion active for each token. It handles text and images, and it was trained on 3,800 Nvidia Grace Blackwell GPUs in Mistral's own datacentres in Europe.

Mistral says it has three reasoning modes: Low for simple tasks, Medium for everyday tasks and High, the default, for the hardest ones.

Who can use it, and what does the licence say?

The files are public on Hugging Face. The licence is the Mistral Ecosystem Open-Weight License, which Mistral describes as a modified version of the MIT licence. It lets anyone use, change, share and sell copies of the model.

There is one limit. A company that offers the model to others as a service, for example through an API, needs a separate commercial licence from Mistral. The licence says this does not apply to internal use, or to use through Mistral's own products.

What do Mistral's benchmarks show?

Mistral reports 41% on Terminal-Bench 4.0, up 13 percentage points since the preview. It says the model leads on Cognition's FrontierCode 2.0 with 19.7%, and scores 51% on the Artificial Analysis Cyber Index.

It also says the model refuses harmful cyber requests 96.6% of the time on average. These are Mistral's own figures. We have not tested them.

Open models let security researchers and developers inspect, adapt and build on AI.

Kari Briski, VP of Open Models and Software at Nvidia

Why does it matter?

Most top models can only be used through a company's servers. A model of this size with public weights lets companies and governments run it themselves, under their own rules. Most people will still reach it through apps and services built on top of it.

What's next?

Developers and companies can now download the model and test it on their own hardware. Independent benchmarks will show how it compares with Mistral's claims.