French AI company Mistral has made available the operational weights for its trillion-parameter open-source AI model, Mistral Large 4 (ML4), empowering businesses and governments to download and deploy the system autonomously. Affectionately termed “Le Chonk,” this model was initially unveiled earlier this month.
This release occurs as the competition intensifies between open-weight AI developers, such as Mistral, and providers of closed models, like OpenAI and Anthropic. The former group contends that businesses and governmental bodies should not be beholden to external AI firms for access to cutting-edge capabilities. This argument may carry particular weight in the cybersecurity domain, where organizations prioritize greater command over their data and systems.
As Mistral’s CEO, Arthur Mensch, articulated, AI should “be a tool for empowerment, not dominance.” With Chonk, Mistral reinforces this stance, claiming its ability to achieve “state-of-the-art performance” at a “fraction of the compute and capital committed by closed labs.”
The public release of the model’s weights—the numerical training data that brings the system to life—ensures that other companies and government agencies can now implement ML4 for their specific objectives.
Mistral states that its model’s performance is comparable to that of a Chinese open-weight competitor, GLM-5.3, which was previously considered among the world’s leading open-weight models, especially concerning cyber operations. However, Mistral’s model utilizes 47 percent fewer tokens per task, rendering it considerably more economical than GLM-5.3.
The company also asserts that the model surpasses all other open models in agentic coding tasks, as evaluated by Artificial Analysis’ Coding Agent Index, a prominent AI benchmarking platform.
“Amid escalating cyber threats, ML4 delivers state-of-the-art sovereign cyber defense capabilities to enterprises and governments globally,” the company declared on Thursday. “Enterprise and public sector institutions can now acquire, deploy, and own frontier AI to detect, analyze, and respond to threats across code and IT infrastructure.”
The weights for ML4 are being made accessible on Hugging Face and through its API.
Pierre Stock, Mistral’s Vice President of Science, informed TechCrunch that the release of the model’s weights grants organizations increased autonomy over how they utilize AI to protect themselves.
“Concretely, this signifies that, for instance, a business entity or a government can gain access to cutting-edge intelligence to defend themselves against cyberattacks. This access cannot be revoked by definition because the model is open-weighted,” he stated.
Stock further indicated that the model could be employed to “identify, coordinate, and respond in real time to significant security incidents,” and that its open-source nature allows companies to deploy this model on-premises within their proprietary ecosystems.
While advancements in AI cybersecurity are empowering security professionals to uncover vulnerabilities at a significantly higher rate than before, Stock continued, he believes it is vital for companies to “furnish [this identical] capability to everyone…as promptly as possible to enhance and fortify the ecosystem,” acknowledging that “some malicious actors might endeavor to actually leverage their own technology to attack.”
Separately, Stock addressed concerns regarding recent events involving rogue AI agents, recognizing the inherent risks but advising against an overreaction.
“We do not intend to incite fear. We believe this is a solvable issue. Achieving robust security poses a substantial challenge,” he remarked.