Mistral AI Releases Open-Weight Mistral Large 4
Mistral AI released the open weights of Mistral Large 4, a 1-trillion-parameter model it says leads open-weight benchmarks in coding and cybersecurity.
Quick answer
What did Mistral AI announce about Mistral Large 4?
Mistral AI released the open weights of Mistral Large 4 (Le Chonk), a 1-trillion-parameter multimodal model. The company says it leads open-weight coding and cybersecurity benchmarks, offers three reasoning modes, and was trained on European infrastructure for sovereign AI deployment.
Key takeaways
- Mistral AI released the open weights of Mistral Large 4, also called Le Chonk, a 1-trillion-parameter, natively multimodal model that had been in public preview earlier in October.
- The company says additional reinforcement learning since the preview improved coding and defensive cybersecurity performance, with Mistral Large 4 taking the lead among open-weight models on Cognition's FrontierCode 2.0 benchmark (19.7%) and Terminal-Bench 4.0 (41%).
- Mistral says the model scores 48% on the Artificial Analysis Coding Agent Index, roughly matching GLM-5.3 while using 47% fewer tokens per task.
- On the Artificial Analysis Cyber Index, Mistral says the model scores 51% for discovering, reproducing and patching vulnerabilities, with a 96.6% average refusal rate on malicious cyber requests.
- Mistral says the model supports three reasoning modes (low, medium, high) so developers can balance performance against inference cost, and was trained on European infrastructure.
Open weights released
Mistral AI has released the open weights of Mistral Large 4, the 1-trillion-parameter, natively multimodal model it calls Le Chonk. The company first put the model into public preview earlier in October and says it has since used developer and partner feedback, along with continued reinforcement learning, to improve the model's coding and defensive cybersecurity performance.
Mistral says the model was trained from scratch on European infrastructure and describes it as a foundation for what it calls sovereign AI, allowing organizations to self-host and customize the model.
Coding performance
Mistral says three additional weeks of reinforcement learning after the preview moved Mistral Large 4 to the front of open-weight coding models. The company cites a score of 19.7% on Cognition's private FrontierCode 2.0 benchmark, which it says evaluates whether a repository maintainer would merge a coding agent's change, placing the model ahead of other open-weight models on that test.
On the Artificial Analysis Coding Agent Index, Mistral says the model scores 48%, which it describes as effectively tied with GLM-5.3, while using 47% fewer tokens per task on average. On Terminal-Bench 4.0, Mistral says the model improved 13 percentage points over its preview version to 41%, taking the lead among open-weight models while cutting output tokens by 30%.
Mistral also points to blind pairwise evaluations run by Surge Code, in which it says Mistral Large 4 was preferred over GLM-5.3, Kimi K3, and DeepSeek V4.1 Flash, ranking behind only Claude Opus 5.5 among the models tested.
Reasoning modes
The company says Mistral Large 4 now supports three reasoning modes that let developers trade off performance against cost: low for simple tasks, medium for everyday tasks, and high, the default, for tasks requiring high precision and thorough reasoning.
Cybersecurity capabilities
Mistral says it worked with cybersecurity vendors to evaluate the model on vulnerability reproduction and patching in real-world environments. The company reports a score of 51% on the Artificial Analysis Cyber Index, which it says measures a model's ability to discover, reproduce, validate and patch vulnerabilities without breaking existing functionality.
Mistral says the model's guardrails distinguish defensive from malicious cybersecurity requests, and that it achieved a 96.6% average refusal rate on malicious cyber requests across JailbreakBench, StrongREJECT and AgentHarm, which the company says is the highest it has measured among open models.
The company says it monitored cybersecurity-related use of the preview model in live traffic and observed the model refusing black-box attacks against live targets, while also identifying a potential for the model to be used to uplift real attacks. Mistral says it is modifying its guardrails in response and will continue to update them.
Mistral says it is giving enterprise design partners, including Orange Cyberdefense, preview access to a specialized variant of the model optimized for penetration testing, vulnerability assessment and red-teaming, with access subject to vetting and contractual safeguards restricting testing to authorized systems. The company says customers can also tailor the model using proprietary security data through a tool it calls Mistral Forge, and can further optimize it into smaller, task-specific variants.
Source: Mistral AI, "Mistral Large 4 (Le Chonk) is now open weight," published October 29, 2026.
Frequently asked questions
- What is Mistral Large 4?
- Mistral Large 4, nicknamed Le Chonk, is a 1-trillion-parameter, natively multimodal model from Mistral AI that the company first released in public preview and has now made available as open weights.
- What coding benchmarks does Mistral cite for Large 4?
- Mistral cites a 19.7% score on Cognition's FrontierCode 2.0 benchmark, a 41% score on Terminal-Bench 4.0, and a 48% score on the Artificial Analysis Coding Agent Index, each described as leading or matching other open-weight models.
- What does Mistral say about the model's cybersecurity performance?
- Mistral says the model scores 51% on the Artificial Analysis Cyber Index for vulnerability discovery, reproduction and patching, and has a 96.6% average refusal rate on malicious cyber requests.
- What reasoning modes does Mistral Large 4 offer?
- Mistral says the model supports three modes: low for simple tasks, medium for everyday tasks, and high, the default, for the hardest tasks requiring precise, thorough reasoning.