Mistral opens Large 4 weights to challenge US and Chinese cyber AI

Mistral released the downloadable weights for its Large 4 cybersecurity AI model on October 29, letting businesses and governments run and customize it in-house instead of relying on American or Chinese providers. The model had been offered through an API since October 6 and remains available that way; reselling it as a hosted service requires a separate commercial license. Mistral lists 1.05 trillion parameters, including 52 billion active, and executive Pierre Stock said access cannot be cut off. On the Artificial Analysis Cyber Index, which measures finding and fixing vulnerabilities rather than conducting attacks, Large 4 scored 51, behind Kimi K3.1 Pro at 64.