Introducing Mistral Large 4
mistral ·
Mistral has launched a public preview of Mistral Large 4, also known as 'le Chonk,' a trillion-parameter natively multimodal model with 49 billion active parameters. Trained from scratch on NVIDIA Grace Blackwell GPUs in European datacenters, the model aims to advance AI sovereignty by providing state-of-the-art performance among open-weight models. It demonstrates exceptional capabilities in cybersecurity, coding, agentic workflows, and financial or legal tasks, often surpassing leading closed-source competitors like GPT-6 Astra. Notably, ML4 excels in visual grounding and resists prompt injection attacks more effectively than other open-source alternatives. The company plans to release the model's weights by the end of October after completing red-teaming exercises with security partners. This launch is supported by Mistral’s €3 billion Series D funding, which is being used to scale infrastructure and continue reinforcement learning training. The model is fluent in over 160 languages and is designed for self-deployment to ensure autonomy for enterprise clients.
mistral