Mistral AI, a French artificial intelligence startup, has announced the public preview of its latest flagship open-weight model, Mistral Large 4.
Nicknamed “le Chonk” by its company, this model marks a major expansion in open-weight artificial intelligence capabilities. Built as a 1 trillion-parameter natively multimodal model with 49 billion active parameters, the French AI company claims that it is their largest and most capable system to date. They share that they trained their model on 3,800 NVIDIA Grace Blackwell GPUs across Mistral’s sovereign European data centers. Its multilingual training dataset spans over 160 languages, covering every official language of the European Union, and the model operates under European law to offer full digital sovereignty to enterprise clients.
ML4 represents the first major milestone funded by Mistral’s €3 billion Series D investment, making it the largest equity round ever raised by a European technology company. The company shares that its underlying architecture and reinforcement learning setup will serve as the base for a new generation of industry-specialized models.
Their previous release was Mistral Small 4. While Mistral Small 4 and Mistral Large 4 are two similar pillars of Mistral AI’s 4th-generation model family. They share the same underlying core technology and generation design, but they are built for entirely different deployment targets, scales, and operational roles.
