Mistral’s new “Large 4” model bets on cybersecurity strengths over raw IQ

European AI startup Mistral has rolled out its latest flagship model, Mistral Large 4—nicknamed “Le Chonk”—positioning it as a serious contender in the next wave of foundation models. Independent testing by benchmarking lab Artificial Analysis suggests the model’s overall “intelligence index” still trails several cutting‑edge Chinese systems, but tells a different story in one critical area: cybersecurity.

On tasks such as analysing suspicious code, surfacing potential vulnerabilities and reviewing configuration for security gaps, Large 4 ranks among the top performers. Mistral is leaning into this profile, arguing that the AI race is no longer just about raw scores but about which models can be safely controlled and adapted to specific enterprise needs.

Large 4 is released as an open‑weight model, meaning organisations can host and fine‑tune the underlying parameters themselves. That makes it attractive for building in‑house defence and audit agents tuned to sensitive environments. At the same time, open‑weight models carry a dual‑use risk, as attackers can also probe and repurpose them. This leaves Mistral and other Western labs trying to strike a balance: challenging China’s growing dominance in open models while helping shape norms and safeguards around how such systems are shared and governed.