Introducing Mistral Large 4

Mistral has launched a public preview of Mistral Large 4, nicknamed "le Chonk", a natively multimodal model with 1 trillion total parameters and 49 billion active parameters, making it the company's largest and most capable model to date. It achieves frontier-level performance competitive with the strongest open-source models and outperforms open-weight models from the US and Europe, with particular strength in cybersecurity, finance, law, coding, agentic workflows, and multimodal understanding, and weights are due for release by the end of the month. The model was trained from scratch on 3,800 NVIDIA Grace Blackwell GPUs in Mistral's own European datacenters, positioning it as a milestone for European AI sovereignty and giving organizations open-weight, self-deployable control over a model that leads on cyber benchmarks such as the Artificial Analysis Cyber Index and Cybench, where closed models often refuse tasks. Additional capabilities span agentic coding, business workflow automation, scientific and mathematical reasoning, and knowledge work like spreadsheets, slides, and legal and financial analysis, all supported by a large-scale reinforcement learning pipeline running on roughly 3,000 GPUs that produces about 33 billion tokens per day. The work is funded by Mistral's €3 billion Series D, the largest equity round raised by a European technology company, and the model will serve as the foundation for a new generation of specialized Mistral models.

(Source：Mistral)