On October 29 local time, French company Mistral AI officially released the model weights of Mistral Large 4 (ML4 for short, codename le Chonk). The model is a natively multimodal model with 1 trillion total parameters, 49 billion activated parameters, and 52 billion parameters including embedding and output layers. It was trained from scratch by Mistral in its own European data center using 3,800 NVIDIA Grace Blackwell GPUs. This version was performance-optimized and upgraded from the initial version announced three weeks ago, following real-world testing with government and enterprise users. It scored 48% on the Artificial Analysis Coding Agent Index, roughly on par with leading open-source model GLM-5.3, while reducing average token consumption per task by 47%. In terms of pricing, input is $0.70 per million tokens, cached input is $0.07 per million tokens, and output is $2.10 per million tokens. The model is released under a modified MIT license and can be used free of charge for personal, academic, modification, fine-tuning, and conventional commercial use; cloud service providers or vendors offering MaaS resale or commercial use must separately obtain commercial authorization from Mistral.