Mistral AI has introduced its new flagship model, Mistral Large 4, also known as Le Chonk. The system features 1 trillion parameters, with 49 billion active at any given time, and utilizes a Mixture-of-Experts architecture. The model is natively multimodal and supports over 160 languages, including all official European Union languages.
The training process for the model lasted approximately two months and required 4,000 NVIDIA Grace Blackwell GPUs located in European data centers. On the Artificial Analysis Intelligence Index, the model achieved a score of 38 points. For comparison, the Chinese model MiMo-V2.6-Pro scored 46 points on the same index.
Mistral is positioning the model for use in cybersecurity, finance, coding, manufacturing, and visual grounding. Currently, the system is in limited preview for select developers. The company plans to release the model weights to the public on October 27, 2026.
For users accessing the model via API, Mistral has set pricing at $1.36 per million tokens for input and $4.18 per million tokens for output.
