French AI startup Mistral has made the weights for its massive trillion-parameter model, called Mistral Large 4 (ML4) or “Le Chonk,” publicly available, enabling businesses and governments to download, deploy, and fully own the system themselves. Introduced earlier in October 2026, ML4 is being pitched as a more transparent, controllable frontier AI compared to closed-source rivals.
This move comes amid heightened tension between open‐weight AI developers, which favor releasing model internals, and closed labs like OpenAI and Anthropic that retain control of their weights. Mistral argues that when it comes to cybersecurity—a realm where sensitive infrastructures and classification demand full oversight—open weights are essential.
What ML4 Offers
ML4 is designed to deliver “state-of-the-art performance” in cybersecurity tasks while demanding significantly less computational and capital investment than many closed models. The company points out that ML4 matches or beats GLM-5.3, a strong open-weight Chinese model, while using about 47% fewer tokens per task—making operations cheaper and more efficient.
Beyond matching GLM-5.3 in general performance, ML4 reportedly leads among open models in agentic coding tasks, according to rankings on the Artificial Analysis Coding Agent Index. This suggests strong proficiency in writing and understanding code—skills crucial for detecting bugs, automating defensive programming, or analyzing exploits.
Why Open Weights Matter in Cybersecurity
By releasing the model’s weights on platforms like Hugging Face and exposing ML4 via its own API, Mistral makes it possible for organizations to deploy the model entirely on their own infrastructure. This is especially relevant for enterprise and government users who require full sovereignty over data, runtime, and security.
ML4 is meant to help with identifying, analyzing, and responding to cybersecurity threats—ranging from code vulnerabilities to large security incidents—within an organization’s own environment. Because ML4 is open‐weight, its users are less likely to be impacted by restrictions or cut-offs imposed by third-party service providers.
Mistral emphasizes that while the expansion of AI agents poses risks (e.g. rogue agents), these challenges are solvable rather than reason for alarm. The goal, from its perspective, is to strengthen defenses broadly, not just build capabilities for isolated actors.
ML4’s availability marks an important milestone in the race between closed and open AI models. Organizations now have access to a frontier model they can run and secure themselves, without depending on external providers for updates or continuity.
