Mistral Launches 1-Trillion-Parameter Mistral Large 4, Claiming Open-Weight Flagship Rivals Best Closed Models

Mistral has launched a public preview of Mistral Large 4, a 1-trillion-parameter open-weight flagship model — nicknamed le Chonk and built for general agentic capabilities — that the company claims rivals the best closed models while competing with open-weight models three times its size; trained on 4,000 NVIDIA Grace Blackwell GPUs with reinforcement learning, the model runs on Mistral's European data centers, targets on-premise and private-cloud enterprise deployments, and will see its weights fully released on October 27.

• Mistral on Tuesday launched a public preview of Mistral Large 4 (ML4), le Chonk, its new flagship 1-trillion-parameter open-weight model built for general agentic capabilities, claiming it rivals the best closed models while competing with open-weight models three times its size.
• The model was trained on 4,000 NVIDIA Grace Blackwell GPUs at what Mistral describes as a fraction of competitors' cost, and co-founder and chief scientist Guillaume Lample says its open-weight nature lets companies build cybersecurity defenses without fear the model will be deprecated or sunset.
• The weights will be fully released on October 27; the model runs on Mistral's own European data centers, is available via its API, targets on-premise and private-cloud enterprise deployments, and was trained with reinforcement learning rather than the distillation used by many open models.