Mistral AI launches ml4, a 1 trillion-parameter open-source model with european deployment

On October 6, 2026, Mistral AI, a French company, launched the public preview version of its Mistral Large 4 (ML4) model, codenamed le Chonk. The model, with 1 trillion total parameters and 490 billion activation parameters, is available for testing via Mistral Studio, with weights set to be released by the end of the month. Mistral claims ML4 outperforms all open-source models developed in Europe and the U.S., achieving state-of-the-art results in critical enterprise applications like cybersecurity, finance, and law. The model was trained entirely within Mistral’s European data center using 3,800 NVIDIA Grace Blackwell GPUs, adhering strictly to EU regulations. It supports over 160 languages, including all official EU languages. Pricing starts at $1.36 per million input tokens and $4.18 per million output tokens. Mistral is also conducting red-team testing with trusted partners and regulators to ensure security and compliance.