Mistral has released a trillion-parameter model for cybersecurity

French company Mistral has released the weights for its new ultra-large model Mistral-Large-4-1T-A52B, which features a trillion parameters (52 billion of which are active). The release is focused on cybersecurity tasks.

The creators position the development as a tool capable of providing organizations with a fundamentally new level of control over cyber defense. Unlike closed proprietary systems, the open release allows for deploying AI directly within a company's private infrastructure.

The new model is designed to help security professionals automate vulnerability detection, analyze network traffic, and respond promptly to incidents. The architecture is based on a Mixture-of-Experts approach, supports a 1-million-token context window, works with over 160 languages, and demonstrates strong results in programming and logical reasoning. The open architecture makes it possible to tailor the system to the specific requirements of corporate systems without transferring data to third-party service providers.

The release of the model provides large enterprises with a tool for creating sovereign threat analysis systems with deep customization for local security standards.