The French AI company Mistral has released the weights of its trillion-parameter open-source AI model, Mistral Large 4 (ML4), enabling businesses and governments to download and deploy the system themselves. The model, nicknamed "Le Chonk," was first unveiled earlier this month.
This release comes as competition intensifies between open-weight AI developers, such as Mistral, and closed-source model providers, such as OpenAI and Anthropic. The former argue that businesses and governments should not rely on external AI companies to gain advanced capabilities. This argument carries particular weight in the cybersecurity field, where organizations seek greater control over their own data and systems.
As Mistral’s CEO Arthur Mensch said, AI should “be an empowering tool, not a dominating one.” With Chonk, Mistral further reinforces this stance, claiming it can achieve “industry-leading performance” with just a fraction of the compute and capital invested by closed-source labs.
The release of model weights—the numerical training results that drive the system—means that other companies and government agencies can now deploy ML4 for their own purposes.
Mistral stated that, in terms of performance, its model is comparable to China’s open-weight competitor GLM-5.3, which was previously regarded as one of the world’s top open-weight models, especially for cybersecurity tasks. However, Mistral’s model uses 47% fewer tokens per task, making it significantly more cost-efficient than GLM-5.3.
The company also stated that, on the Coding Agent Index from the AI benchmarking website Artificial Analysis, this model outperformed all other open models on agent programming tasks.
Mistral said on Thursday: "As cyber threats continue to rise, ML4 delivers cutting-edge sovereign cyber defense capabilities to global enterprises and governments. Enterprises and public sector organizations can now download, deploy, and own state-of-the-art AI to detect, analyze, and respond to threats in code and IT infrastructure."
The weights for ML4 will be available through Hugging Face and its API.
Pierre Stock, Vice President of Science at Mistral, told TechCrunch that releasing model weights will give various organizations greater control over how they defend against AI.
He said, “Specifically, this means that, as a business or a government, you will be able to access cutting-edge intelligence to protect yourself from cyberattacks. This access cannot be cut off by definition, because the model is open-weight.”
Stock also stated that the model can be used to "real-time identify, coordinate, and respond to major security incidents," and that open-sourcing it allows enterprises to deploy the model locally within their own ecosystems.
Stock continued to say that although advancements in AI for cybersecurity are helping security professionals discover vulnerabilities at a much faster rate than before, he believes it is crucial to provide this capability to everyone as soon as possible—"to accelerate the enhancement and strengthening of the entire ecosystem," given that "some malicious actors may indeed attempt to use their own technologies to launch attacks."
Additionally, Stock addressed concerns from the public regarding the recent out-of-control AI agent incident. He acknowledged the risks but cautioned against overreacting.
He said, "We don't want to cause panic. We believe this is a solvable issue. Ensuring security is a challenging task."