Mistral opens the weights to its 1-trillion-parameter Large 4
Le Chonk moves from API-only preview to self-hostable weights on Hugging Face, under a licence that still requires a separate deal if you resell it as a hosted service.
Mistral released the open weights for Large 4, nicknamed Le Chonk, on 29 October. The 1-trillion-parameter mixture-of-experts model had only been available as an API preview until now. The weights are on Hugging Face under the mistralai/Mistral-Large-4-1T-A52B repo, with deployment support for vLLM, SGLang and Nvidia’s NIM container.
The licence, the Mistral Ecosystem Open-Weight License, is a modified MIT: free to use, modify, fine-tune and sell as a product, but a business running its own Model-as-a-Service offering needs a separate commercial licence from Mistral first. Mistral says the preview-to-release gap bought three more weeks of reinforcement learning, moving Terminal-Bench 4.0 up 13 points to 41%, and claims the model beats GLM-5.3 on Artificial Analysis’s Coding Agent Index using 47% fewer tokens. API pricing stays at $0.70 per million input tokens and $2.10 per million output tokens.
Why it matters: this is now a frontier-sized open model you can run on your own hardware instead of renting through an API, which matters if data residency or vendor lock-in is the actual constraint. The catch is in the fine print: fine-tuning and shipping it inside your own product is fine, but standing it up as a hosted inference service for others is not, without talking to Mistral first.