Mistral Large 4 Release: 1-Trillion-Parameter Open-Weight Model Details

Mistral AI has officially released the weights of its flagship multimodal model, Mistral Large 4, also known as Le Chonk. Featuring 1 trillion parameters, the model is trained entirely on European infrastructure and delivers enhanced reasoning, coding, and defensive cybersecurity capabilities.

• Mistral Large 4 is a flagship 1-trillion-parameter multimodal model with fully open weights.
• It leads open-weight benchmarks in coding while utilizing significantly fewer tokens.
• Features robust defensive cybersecurity capabilities with a 96.6% refusal rate for malicious requests.
• Offers three reasoning modes to balance performance and inference costs effectively.

Breakthrough Coding Performance

Following weeks of additional reinforcement learning, Mistral Large 4 has achieved top-tier performance in software development tasks. On Cognition's private FrontierCode 2.0 benchmark, it leads with 19.7%. On the Artificial Analysis Coding Agent Index, it scores 48%, tying with the leading open-weight model GLM-5.3 while using 47% fewer tokens per task.

In terminal workflows, Terminal-Bench 4.0 results improved by 13 percentage points to 41%, reducing output tokens by 30%. In Surge Code blind evaluations, Mistral Large 4 ranked second only to Claude Opus 5.5 among evaluated models. Developers can now leverage three distinct reasoning modes to optimize the trade-off between execution speed, precision, and cost.

Advanced Cybersecurity and Defensive Tooling

The model excels at vulnerability discovery, reproduction, and patching without disrupting existing functionality. It scores 51% on the Artificial Analysis Cyber Index. Built-in safety guardrails successfully differentiate between defensive remediation and malicious cyber activities.

Mistral Large 4 achieves an average refusal rate of 96.6% on malicious cyber requests, establishing a high benchmark for open safety. Industry leaders such as Booz Allen Hamilton, Cisco, CrowdStrike, and Elastic have evaluated the model, highlighting its value for data-sovereign deployments and customized threat intelligence operations.

Significance for Enterprises and Sovereign AI

The open-weight release of Mistral Large 4 defines a major milestone for sovereign AI, enabling organizations to own, customize, and control their infrastructure. Enterprises and institutions can deploy the model locally, maintaining strict adherence to data residency and privacy mandates.

By combining extreme token efficiency with deep reasoning and defensive security skills, Mistral Large 4 empowers engineering teams to build resilient architectures. Organizations can run sophisticated autonomous coding agents and cybersecurity pipelines securely on their own hardware without relying on third-party cloud APIs.