We're releasing the weights of Mistral Large 4

The weights for a 1-trillion-parameter multimodal model are now publicly available, enabling greater customization and self-hosting.

Mistral AI has released the weights for its Mistral Large 4 (Le Chonk) model, a 1-trillion-parameter multimodal model. This release follows a public preview and incorporates significant improvements based on user feedback, particularly in coding and cybersecurity capabilities. Mistral Large 4 now leads open-weight benchmarks for coding tasks, demonstrating enhanced efficiency and performance on benchmarks like FrontierCode and Terminal-Bench. It also shows strong capabilities in cybersecurity, excelling at vulnerability detection and patching while maintaining a low refusal rate for malicious requests. The model offers three reasoning modes to balance performance and cost, and its European training and open-weight nature are highlighted as key factors for sovereign AI deployments. Mistral AI emphasized its commitment to releasing capable models openly to empower users with control and customization.

Mistral Large 4's release signifies a major advancement in open-weight AI, pushing the frontier for agentic coding and defensive cybersecurity applications. Its leading performance on coding benchmarks, coupled with superior token efficiency, makes advanced coding agents more accessible and economical for widespread use. The model's enhanced cybersecurity capabilities, validated by industry leaders, offer organizations a powerful tool for vulnerability management and threat intelligence, crucially without the risk of unintended refusals on legitimate security tasks. The availability of three distinct reasoning modes provides developers with granular control over the trade-off between computational cost and performance, allowing for tailored deployments across a spectrum of use cases, from simple tasks to highly complex, precision-intensive operations.

This release directly addresses the growing demand for Sovereign AI solutions, offering organizations a path to self-hosted, customizable AI with full data control. Mistral AI's commitment to transparency and open development positions them as a key player in democratizing access to frontier AI technology. The model's strong performance and adaptability are likely to accelerate innovation in sectors requiring secure and controlled AI deployments, such as finance, critical infrastructure, and government. As adversarial AI tactics evolve, the availability of robust, auditable, and customizable defensive AI models like Mistral Large 4 becomes increasingly vital for maintaining security parity.