Mistral Large 4: "Le Chonk"

Mistral Large 4, unofficially known as "ML4" or "le Chonk," has been launched with a public preview available on Mistral Studio. This model, with 1 trillion parameters and 49 active parameters, is the largest and most capable model unveiled by Mistral thus far. It demonstrates exceptional performance across various domains, including coding, agentic workflows, and multimodal understanding.

ML4 has already achieved performance levels comparable to the best open-source models globally and surpasses open-weight models developed in the US or Europe in several critical enterprise workloads, such as cybersecurity, finance, and law. In fact, ML4 is currently considered state-of-the-art among open models in these sectors.

To ensure the model's robustness, it is being red-teamed with cybersecurity leaders, vetted partners, and state authorities who will access ML4 with reduced moderation and expanded cyber capabilities. The model was trained on 3,800 NVIDIA Grace Blackwell GPUs situated in Mistral's European datacenters. The public preview of ML4 is served on the same infrastructure, highlighting Mistral's investment in infrastructure, research, and product development to deliver state-of-the-art performance in crucial verticals through open weights.

The model will be accessible in multiple regions worldwide, including a European deployment managed independently by Mistral under European law. Notably, a significant portion of ML4's training data is multilingual, encompassing over 160 languages, including every official language of the European Union.