Mistral has opened its new Le Chonk model, officially called Mistral Large 4 (ML4), for public preview. The company says the trillion‑parameter model is built for cybersecurity, letting enterprises keep full control over the model and their data. It will release the model weights on Oct. 27, after a month of testing with developers and security teams. Mistral trained ML4 on 4,000 Nvidia Grace Blackwell GPUs over two months, a fraction of the compute used for OpenAI’s Astra model.
The lab also promises flexible deployment, including self‑hosting or API access in a European sovereign region, so data stays under EU jurisdiction. Mistral claims ML4 outperforms comparable open models from Kimi, DeepSeek and Meta in cyber capabilities.
Why it matters
Enterprises can adopt a high‑capacity AI defender without depending on a closed‑source vendor that could revoke access.
