Mistral Releases Weights of Mistral Large 4, a 1T-Parameter Open Model with a Model-as-a-Service Carve-Out
The European lab's trillion-parameter MoE is now downloadable in FP8 and NVFP4, though independent benchmarks place it behind leading Chinese open models.
Mistral AI released the weights of Mistral Large 4 — nicknamed "Le Chonk" — on October 29, three weeks after the model's API preview on October 6. The multimodal mixture-of-experts model carries roughly 1.05 trillion parameters with 49B active per token across 128 experts plus one shared expert and a Pixtral 1.6B image encoder, a 1M-token context window, and was trained on 3,800 Grace Blackwell GPUs in Europe. Mistral says added reinforcement learning lifted its Terminal-Bench 4.0 score from 28% to 41%.
Independent readings are cooler than the vendor framing: per Artificial Analysis figures cited by The Decoder, the model's Intelligence Index stands at 26, behind DeepSeek V4.1 Flash and GLM-5.3, both at 32 — meaning its "top open model" claim holds outside China rather than overall.
The weights ship in FP8 and NVFP4 under the Mistral Ecosystem Open-Weight license, a MIT derivative permitting commercial use and fine-tuning but requiring a separate license to sell the model as a service.