Mistral Large 4: A Trillion Parameters and Cybersecurity

Mistral AI, the French AI lab, released Large 4, an open-weight model with a trillion parameters. Only 49 billion of those activate per token. It is a frontier-level model, multimodal, and strong at code, agentic tasks, finance, and law. The company lists cybersecurity as one of its strong points. In one specific test, reproducing and patching a real vulnerability in open-source code, it scored 82 percent. That beat every other model tested. Claude Opus 5.5 and GPT-6 Astra scored close to zero on the same test, because they refuse to do this kind of work. The API is already open. Pricing is $1.36 per million input tokens and $4.18 per million output tokens.

A model willing to take on security tasks that more restricted models refuse is a real, useful difference, not a marketing trick. Penetration testers can use it right now. For everyday coding work, there is no reason yet to move away from Opus and Fable. Large 4 fills its own niche. It does not replace the main tool.