Mistral Opens Weights of Trillion-Parameter Model for Cyber Defense

Mistral has released the weights of its trillion-parameter AI model, letting businesses and governments download, deploy and own a frontier system on their own servers. The model, Mistral Large 4 (ML4), carries the internal nickname "Le Chonk" and was first introduced earlier this month.

The move sharpens the fight between open-weight developers like the French company and closed-model providers such as OpenAI and Anthropic. Open-weight backers argue that businesses and governments should not depend on outside AI companies for advanced capabilities. That argument carries particular force in cybersecurity, where organizations demand tighter control over their data and systems.

Mistral's CEO, Arthur Mensch, framed the philosophy plainly: AI should "be a tool for empowerment, not dominance." With ML4, Mistral doubles down on that position. The company claims it achieves "state-of-the-art performance" at a "fraction of the compute and capital committed by closed labs."

What does the release actually change?

Releasing the weights — the numerical parameters produced by training that animate the system — means other companies and government agencies can now run ML4 for their own purposes. Mistral made the weights available Thursday on Hugging Face and via its API.

"Amid growing cyber threats, ML4 brings state-of-the-art sovereign cyber defense capabilities to enterprises and governments globally," the company said. "Enterprise and public sector institutions can now download, deploy, and own frontier AI to detect, analyze, and respond to threats across code and IT infrastructure."

Pierre Stock, Vice President of Science at Mistral, told TechCrunch that open weights give organizations control over how they use AI defensively.

"Concretely, it means that, for instance, as a business company or as a government, you're able to have access to frontier intelligence to defend yourselves against cyberattacks. Access that can't be cut off by definition because the model is open-weighted," he said.

Stock added that the model can "identify, coordinate, and respond in real time to major security incidents." Open sourcing also lets companies deploy the model on-premises, inside their own ecosystem, with no external dependency.

How does ML4 compare with rivals?

Mistral says the model matches the performance of GLM-5.3, a Chinese open-weight competitor previously regarded as one of the best open-weight models in the world, especially for cyber work. But ML4 uses 47 percent fewer tokens per task, which makes it significantly more economical to run than GLM-5.3.

The company also claims the model outranks all other open models on agentic coding tasks measured by Artificial Analysis' Coding Agent Index, an independent AI benchmarking site.

What about rogue AI agents?

Stock addressed recent incidents involving rogue AI agents. He acknowledged the risks but cautioned against overreacting.

"We don't want to fearmonger people. We think it's a problem that is solvable. It's a hard problem to have good security," he said.

His broader argument rests on speed. AI advances are already helping security professionals find bugs at a much higher rate than before. Companies should "provide [this same] capability to everyone…as soon as possible to ramp up and to strengthen the ecosystem," Stock said, because "some malicious actors might try to actually leverage their own technology to attack."

The calculation is straightforward. If offensive tools spread faster than defensive ones, defenders lose. Mistral is betting that sovereignty — the ability to own the model, host it internally and never lose access — will decide which AI providers win security-conscious customers in the months ahead.