Mistral AI has officially released the weights for Mistral Large 4, a flagship 1-trillion-parameter multimodal model internally nicknamed "Le Chonk." The release follows a public preview period during which the company gathered developer feedback to refine the model's performance. The new weights are designed to excel in agentic coding and defensive cybersecurity, positioning the model as a high-capacity option for organizations seeking to own, customize, and self-host their AI infrastructure.
This image represents the internal nickname 'Le Chonk' for the Mistral Large 4 model mentioned in the first paragraph.
The model's technical improvements center on reasoning efficiency and specialized task performance. In coding benchmarks, Mistral Large 4 has taken the lead on the FrontierCode 2.0 and Terminal-Bench 4.0 benchmarks. Notably, it achieved a score of 48% on the Artificial Analysis Coding Agent Index, matching the leading open-weight model (GLM-5.3) while utilizing 47% fewer tokens per task. To balance performance and cost, the model supports three distinct reasoning modes: Low (for simple tasks), Medium (for everyday tasks), and High (the default for complex use cases requiring high precision).
Beyond coding, Mistral Large 4 is tailored for defensive security. It scored 51% on the Artificial Analysis Cyber Index, which evaluates a model's ability to discover, reproduce, and patch vulnerabilities in a codebase. Mistral AI claims the model maintains a 96.6% refusal rate on malicious cyber requests—the highest among open models—while remaining capable of performing legitimate defensive tasks. However, the release is not entirely "open source" in the traditional sense; the licensing terms require users who operate the model as a service to enter into a separate commercial license with Mistral AI. This distinction between "open weights" and "open source" remains a point of discussion regarding the level of freedom granted to developers compared to OSI-approved licenses.