French AI lab Mistral has officially released the weights for its Large 4 model, known as Le Chonk, demonstrating competitive performance among open AI coding agents with enhanced cybersecurity capabilities.
- Le Chonk ranks top among open coding AI models with a 41% benchmark score.
- Model offers three reasoning levels to tackle varying task complexities.
- Focus on cybersecurity enables enhanced vulnerability identification and defense.
What happened
Mistral has publicly released the weights for its Large 4 AI model, nicknamed Le Chonk, following a preview period and tests with select partners. The model, boasting a trillion parameters, achieved a 41% score on the Terminal-Bench 4.0 coding benchmark, outperforming competing open models such as Opencode’s GLM-5.3 by a narrow margin. It also introduces three reasoning modes—low, medium, and high—to optimize for the complexity of different coding tasks.
This release marks a milestone for Western open-source AI, demonstrating significant improvements in efficiency and performance. Notably, Le Chonk uses 47% fewer tokens per task compared to some peers, highlighting its resource efficiency. In overall composite benchmarking, it scores slightly behind GLM-5.3 and on par with Codex GPT-6.1 Terra, reaffirming its position among top AI coding models available in open access.
Why it matters
Le Chonk’s release emphasizes a growing trend where open AI models are not only catching up to but rivaling proprietary models in capability, efficiency, and security focus. As large proprietary AI systems remain costly and sometimes restricted by vendor controls, economical open models like Le Chonk provide more accessible and customizable alternatives for developers and organizations.
Security is a core design goal for Le Chonk, addressing rising concerns about AI model vulnerabilities and misuse. The model scored well on CyberGym cybersecurity benchmarks, adept at identifying and patching real software vulnerabilities, which is critical as AI-driven cyberattacks become more sophisticated. Mistral’s commitment to user control and sovereignty over AI deployments contrasts with closed systems that can revoke access unexpectedly, positioning open models as vital tools in corporate and governmental cyber defense.
What to watch next
Industry observers should track how Mistral and its competitors evolve open AI models to balance cutting-edge coding abilities with strong security safeguards. The reception and adoption by developers, especially in cybersecurity-sensitive environments, will be key indicators of open models’ practical viability.
Additionally, the involvement of companies like Mistral in initiatives such as Nvidia’s Open Secure AI Alliance signals increasing collaboration to democratize AI safety tools. This collective effort aims to counteract frequent AI security incidents and may further boost trust and reliance on open AI frameworks in commercial and government applications.