Mistral releases the weights of Mistral Large 4, its trillion-parameter model, with a focus on cybersecurity

Mistral, the French AI company, has released the weights of Mistral Large 4 (ML4), a one-trillion-parameter open-source model nicknamed "Le Chonk." According to TechCrunch, the model was unveiled earlier this month. With the weights released, companies and governments can download and deploy it themselves. The weights are available on Hugging Face and through Mistral's API.

The launch comes amid growing competition between open-weight model developers, such as Mistral, and closed-model providers, such as OpenAI and Anthropic. The former argue that companies and governments should not have to depend on outside firms to access advanced AI capabilities. The article notes that this argument may carry more weight in cybersecurity, where organizations want more control over their data and systems. Mistral CEO Arthur Mensch summed it up by saying AI must be "a tool for empowerment, not domination."

Mistral claims ML4 achieves "state-of-the-art" performance with "a fraction of the compute and capital" committed by closed labs. According to the company, its performance is comparable to that of GLM-5.3, a Chinese open-weight model regarded until now as one of the best in the world, particularly on cybersecurity tasks. The difference lies in efficiency: Mistral says ML4 uses 47% fewer tokens per task, which would make it considerably cheaper. The company also claims it outperforms all other open models on agentic coding tasks, according to Artificial Analysis's Coding Agent Index. These are the company's own claims, and the article does not indicate that they have been independently verified.

In its announcement, Mistral presents ML4 as a "sovereign cyber defense" capability. It says enterprise and public-sector institutions can download, deploy and own frontier AI to detect, analyze and respond to threats in code and IT infrastructure.

Pierre Stock, Mistral's VP of Science, told TechCrunch that releasing the weights gives organizations more control over how they use AI to defend themselves. According to him, a company or government gets access to frontier intelligence against cyberattacks, access that "by definition" cannot be cut off because the model is open-weight. He added that the model could be used to identify, coordinate and respond in real time to major security incidents. Because it is open, companies can deploy it on their own premises (on-prem).

Stock acknowledged that advances in AI-driven cybersecurity are already helping professionals find bugs at a much faster pace. That is why he argued for giving that same capability to everyone as soon as possible, to strengthen the ecosystem, given that some malicious actors could try to use the technology to attack. On recent incidents involving rogue AI agents, he acknowledged the risks but urged against overreacting: "We don't want to scare people. We think it's a solvable problem. Getting security right is hard."

The article does not detail the model's architecture, its exact license or independent test results. The performance comparisons should be treated as manufacturer claims until external evaluations are available.