By Hamza Faisal

French AI lab Mistral has published the open weights of Mistral Large 4, the trillion-parameter model it nicknames "Le Chonk", turning a preview that was previously available only through its hosted service into a model anyone can download, run on their own servers and adapt. The weights went live on Hugging Face on October 29, three weeks after the model first appeared in public preview.

The release is the most significant open-weights drop from a European lab to date. Mistral describes Large 4 as its flagship: a natively multimodal mixture-of-experts system with about one trillion parameters in total, of which 49 billion are active for each token — 52 billion counting embeddings and output layers. It accepts text and images, writes text, and carries a context window of one million tokens, with fluency in more than 160 languages, including every official language of the European Union.

What the release actually contains

The weights are listed on Hugging Face as Mistral-Large-4-1T-A52B and are served through Mistral's own API, OpenRouter, Microsoft Foundry — the first cloud to carry it — and Nvidia's NIM container format. Mistral's deployment guidance points builders to the open-source vLLM or SGLang inference engines, and the weight files alone run to 54 safetensors shards, this is not a laptop model: official deployment recipes target multi-GPU servers.

Mistral says the finished version reflects three additional weeks of reinforcement learning and customer testing since the October 6 preview, and it trained the model from scratch in about two months on roughly 4,000 Nvidia Grace Blackwell GPUs in its own European data centres. On benchmarks reported by the company, Large 4 scores 48 percent on the Artificial Analysis Coding Agent Index — effectively tied with China's GLM-5.3, the leading open model there — while using 47 percent fewer tokens per task, and 41 percent on Terminal-Bench 4.0, up from 28 percent at preview. Hugging Face ranks it fifth among open-weights models on the Artificial Analysis Intelligence Index.

The licence catch

The weights ship under Mistral's Ecosystem Open-Weight licence, a modified MIT licence that permits free use, modification, fine-tuning and commercialisation — with one carve-out. Businesses that operate a model-as-a-service offering, renting the model out to others, must obtain a separate commercial licence from Mistral first. The distinction matters for the cloud providers and API resellers that would otherwise simply re-host the model, though ordinary companies building products on top of it face no such requirement.

For those who prefer not to self-host, Mistral's API pricing is 70 cents per million input tokens, 7 cents per million cached input tokens and $2.10 per million output tokens.

A cyber pitch, with limits

Mistral is pitching Large 4 heavily at cyber defenders. The company says the model scores 51 percent on the Artificial Analysis Cyber Index and that in its own tests it refused 96.6 percent of malicious cyber requests. A more aggressive penetration-testing variant, however, stays restricted to vetted partners — a deliberate limit on a model whose capabilities cut both ways.

The timing also sharpens Europe's position in the open-model race. Until now, the strongest open-weights systems have largely come from the United States and China; Large 4 gives European companies and governments a frontier-class model they can host privately, audit themselves and fine-tune without sending data to a foreign provider's servers. Whether developers embrace it at scale will depend less on the headline parameter count than on everyday economics — token efficiency, hosting cost and licence terms — and on that front, Mistral's pitch is that Le Chonk does more with fewer tokens than its closest open rival.