Mistral releases open weights for Mistral Large 4, providing sovereign control for the company’s ‘Le Chonk’ AI model

Mistral AI SAS has unveiled the open weights of its most advanced large language model, Mistral Large 4, also known as "Le Chonk." This 1 trillion parameter model, with 52 billion active parameters, is a fully multimodal AI capable of complex tasks with outstanding performance at a fraction of the cost of competitors. As part of the company's Science team, Pierre Stock highlights Mistral's rapid innovation capacity.

The company's reinforcement learning trained the model on 3,800 Nvidia Grace Blackwell GPUs in their European data centers, ensuring high-quality performance for various applications. Mistral Large 4's open weights allow companies to deploy the model on their own hardware, maintaining sensitive data within their own systems. This development significantly enhances the competition in the AI market, providing a Western alternative to Chinese AI models like GLM from Zhipu Huazhang Technology Co. Ltd. and Kimi from Moonshot AI Technology Co. Ltd. Arthur Mensch, co-founder and CEO, emphasizes open source as crucial to prevent AI concentration among a few actors.

Notably, ML4 outperforms many other models, including Chinese-built GLM-5.3, on the Artificial Analysis Coding Agent Index while using 47% fewer tokens per task on average. With three reasoning modes, Mistral Large 4 allows teams to tailor reasoning depth to the task, optimizing token use per task. The company's release of Mistral Large 4 marks a new era of models encompassing all modalities, following the recent launch of Voxtral 3, a series of voice models and voice agents. Mistral plans to continue expanding its model family in the coming months.