Mistral releases open weights for trillion-parameter Mistral Large 4
Mistral says the 1-trillion-parameter multimodal model now leads open-weight coding benchmarks and posts strong defensive-security results, with three reasoning modes for cost control.
In plain words
Mistral AI has published the model weights of Mistral Large 4, nicknamed Le Chonk, so anyone can download, host and adapt it. The company reports the model performs strongly on coding and defensive-security benchmarks, and offers three reasoning modes that trade capability against inference cost. These results come from Mistral's own announcement and have not been independently verified.
What was announced
Mistral released the weights of Mistral Large 4, its flagship 1-trillion-parameter, natively multimodal model, after a public preview earlier in the month. The company says feedback from the preview informed further reinforcement learning, driving what it describes as massive improvements in coding and defensive cybersecurity, plus gains in reasoning efficiency that lower inference costs.
On benchmarks, Mistral reports the model leads open-weight results on Cognition's FrontierCode 2.0 with 19.7%, scores 48% on the Artificial Analysis Coding Agent Index while using 47% fewer tokens per task than GLM-5.3, and improved 13 percentage points to 41% on Terminal-Bench 4.0. In cybersecurity, it scores 51% on the Artificial Analysis Cyber Index and, per Mistral, refuses 96.6% of malicious cyber requests. The model supports three reasoning modes—low, medium and high—letting developers balance performance against cost. It was trained on European infrastructure, and Mistral positions it for sovereign, self-hosted deployments.
What it does not show
- All benchmark figures and claims come from Mistral's own announcement; the supplied passages show no independent verification of the results.
- The passages are excerpts of the announcement, so full details such as licensing terms, hardware requirements and complete benchmark methodology are not covered.
- Vendor quotes from Booz Allen, Cisco, Cloudflare, CrowdStrike and others are promotional statements published by Mistral, not independent evaluations.
- The specialized cybersecurity variant is limited to vetted enterprise design partners and is not generally available.
For security and IT teams
- Scope
- Mistral Large 4 is a fully open-weight model that organizations can self-host and customize, and Mistral positions it for sovereign AI deployments under the organization's own control.
- Status
- The weights are released now after a public preview; a specialized cybersecurity variant for penetration testing and red-teaming is available only to vetted enterprise design partners under contractual safeguards.
- Access
- Mistral says customers can tailor the model with proprietary security data via Mistral Forge, and distill it into smaller task-specific variants for well-defined cybersecurity tasks.
Sources
Source-support review completed 14 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.
Coverage: Mistral AI: selected passages.
Updates and corrections
No corrections so far.
First published 14 Nov 2026
New to the subject? AI from the Ground Up starts from the beginning.