Mistral Releases 'Le Chonk' Weights for Sovereign Cyber Defense

Mistral has released the weights for its trillion-parameter AI model, Mistral Large 4 (ML4), also known as “Le Chonk.” As TechCrunch first reported, the release allows governments and businesses to download and deploy the system independently, removing reliance on closed-model providers such as Anthropic and OpenAI.

Mistral CEO Arthur Mensch stated that AI should serve as a tool for empowerment rather than dominance. The company claims ML4 provides state-of-the-art performance while requiring a fraction of the capital and compute used by closed labs. Mistral further asserts that ML4 outranks all other open models on the Artificial Analysis’s Coding Agent Index for agentic coding tasks.

In terms of cybersecurity, Mistral positions ML4 as a means for public sector and enterprise institutions to own frontier AI to analyze and respond to threats across IT infrastructure and code. Mistral notes that ML4 is comparable in performance to the Chinese open-weight model GLM-5.3, though ML4 is significantly more economical, utilizing 47 percent fewer tokens per task.

Pierre Stock, Mistral’s Vice President of Science, told TechCrunch that open-weighting ensures access to intelligence that cannot be cut off, enabling organizations to deploy the model on-prem to coordinate responses to security incidents in real time. While Stock acknowledged that malicious actors might use similar technology to launch attacks, he argued that providing these capabilities to the broader ecosystem is necessary to strengthen defenses. Stock also addressed risks regarding rogue AI agents, describing security as a “hard problem” but one that is solvable.