French AI startup Mistral has unveiled Mistral Large 4, a natively multimodal model with one trillion parameters nicknamed "Le Chonk." Currently in public preview with weights scheduled for release on October 27, 2026, the model is positioned as a sovereign European alternative to American and Chinese AI. Mistral is pitching the model's advanced cybersecurity and coding capabilities, highlighting that it performs critical vulnerability testing tasks that closed rivals like Claude and GPT-6 refuse to handle.
Model specifications and release timeline
- ▪Mistral released a public preview of Mistral Large 4, a natively multimodal model with one trillion parameters, on October 6, 2026
- ▪Mistral Large 4, nicknamed "le Chonk," features a fine-grained mixture-of-experts architecture with 1.05 trillion total parameters, a 1.6 billion parameter vision encoder, and a one million token context window
- ▪Mistral plans to release the model weights for Mistral Large 4 on October 27, 2026, after completing safety testing and red-teaming with vetted partners
Performance on benchmark indexes
- ▪Mistral Large 4 scored 38 points on the Artificial Analysis Intelligence Index, a major increase from Mistral Large 3's 9 points, but trailed Claude Opus 5.5 (Max) which scored 58 points
- ▪Mistral Large 4 scored 49.8 percent in the Artificial Analysis Coding Agent Index, placing it ahead of Deepseek V4 Pro and Qwen3.8 Max
Security testing and cyber capabilities
- ▪Mistral co-founder Guillaume Lample stated that Mistral Large 4's cyber defense capabilities will allow enterprises and governments to defend against threat actors who jailbreak closed models
- ▪Mistral Large 4 achieved an 82 percent score on an open-source software vulnerability reproduction and patching test, outperforming Claude Opus 5.5 and GPT-6 Astra which scored near zero due to safety refusals
- ▪Mistral Large 4 blocked 93.3 percent of attacks tested in Lakera's B3 AI Security Benchmark, according to Mistral
- ▪Mistral is red-teaming Mistral Large 4 with security firms, vetted partners, and government agencies, providing them with a version of the model featuring reduced moderation and expanded cyber capabilities
Debatable claims
- ▪Mistral should not release the open weights of its cyber-capable Mistral Large 4 model
- ▪Open-weight AI models are safer for cyber defense than closed-weight alternatives
- ▪Relying on US proprietary AI models poses an unacceptable risk to European technological sovereignty
- ▪Strict safety filters on closed AI models do more harm than good to cybersecurity
Story comments
Loading comments…