French AI company Mistral has published the model weights for its trillion-parameter open‑weight model Mistral Large 4 (ML4), allowing businesses and government bodies to download and deploy the system on their own infrastructure. The model, nicknamed “Le Chonk,” was first introduced earlier this month.
The release comes amid growing competition between open‑weight developers such as Mistral and closed‑model providers like OpenAI and Anthropic. Open‑weight proponents argue that organisations should not have to rely on external AI vendors for access to advanced capabilities — an argument that carries particular weight in cybersecurity, where institutions seek greater control over their data and systems.
Arthur Mensch, Mistral’s CEO, framed the company’s stance by saying AI should be "a tool for empowerment, not dominance." Mistral says ML4 can deliver state‑of‑the‑art performance while requiring a fraction of the compute and capital committed by closed labs.
Performance and efficiency
Mistral compares ML4’s performance to the Chinese open‑weight competitor GLM‑5.3, a model previously regarded among the leading open‑weight systems, especially for cyber tasks. According to Mistral, ML4 uses 47 percent fewer tokens per task than GLM‑5.3, making it materially more economical.
The company also claims ML4 ranks above other open models on agentic coding tasks in the Artificial Analysis Coding Agent Index, an AI benchmarking site.
Cybersecurity applications and on‑prem deployment
Mistral stated that publishing ML4’s weights brings "state‑of‑the‑art sovereign cyber defense capabilities to enterprises and governments globally." The company emphasized that enterprise and public sector institutions can now download, deploy, and own frontier AI to detect, analyse, and respond to threats across code and IT infrastructure.
The ML4 weights are being distributed via Hugging Face and through Mistral’s API, enabling organisations to run the model on‑premise within their own ecosystems.
Pierre Stock, Vice President of Science at Mistral, told TechCrunch that making the weights available will give organisations more control over how they use AI for defence. He noted that because the model is open‑weighted, access "cannot be cut off by definition," which is important for cyber defence.
Stock said the model could be used to identify, coordinate, and respond in real time to major security incidents, and that open‑sourcing allows companies to deploy the model within their own environments. He added that while AI advances help security professionals find bugs at a higher rate, it is important to provide these capabilities widely to strengthen the ecosystem, given that some malicious actors may also leverage technology for attacks.
Addressing concerns about rogue AI agents, Stock acknowledged the risks but cautioned against alarmism: "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."
Implications
By releasing ML4’s weights, Mistral has expanded options for organisations seeking sovereign AI capabilities and on‑prem cybersecurity tools, while reinforcing the broader debate over open versus closed AI models. The move could reduce dependence on external providers for advanced AI functions, but the broader security implications of wider access to powerful models remain a responsibility for developers and the security community.
