Home/New releases/Mistral Large 4

Mistral Large 4

Mistral's 1 trillion parameter multimodal model, nicknamed Le Chonk, now released with open weights.

Mistral published open weights for Mistral Large 4 on October 29, 2026. It is nicknamed Le Chonk. It is a mixture-of-experts model with about 1T total parameters and about 52B active per token (49B not counting embeddings and output layers). The model card lists 128 routed experts, with 4 active per token, plus 1 shared expert. Context is 1 million tokens. It takes text and image input and gives text output. There are three reasoning modes: low, medium, and high. High is the default. Mistral first released it in public preview earlier in October. The October 29 release is an improved version with the weights published. You can use it in Mistral Studio, the Mistral API (model name mistral-large-4), OpenRouter, Microsoft Foundry, NVIDIA hosted endpoints and NIM, and Mistral's Vibe Code. Self-host the weights from Hugging Face with vLLM, which Mistral recommends, or SGLang. An NVFP4 checkpoint for Blackwell GPUs is also on Hugging Face. API price per million tokens: $0.70 input, $0.07 cached input, and $2.10 output. The weights are open under Mistral's own license, the Mistral Ecosystem Open-Weight License, a modified MIT license. They are free to use, modify, fine-tune, and use commercially, except companies running a Model-as-a-Service offering need a separate commercial license from Mistral. Mistral's number: 48% on the Artificial Analysis Coding Agent Index, which Mistral says ties the top open-weight model (GLM-5.3) while using 47% fewer tokens per task on average. At 1T parameters this is a datacenter-scale model, not one for a laptop. Most people should use it through the API or a host. Last checked October 30, 2026.

• Voxtral 3
• Open-weight coding models