Mistral Says Its AI Model ‘Le Chonk’ Has Leapfrogged China in the Open-Weight Race
Reported by Hugging Face, TechCrunch, Wired and 4 more outlets7 sources ↓
The French AI lab claims its latest release matches or outperforms even the best models from DeepSeek and Z.ai in coding and cybersecurity.
Sources
- Hugging FaceRead ↗Welcome Le Chonk: Mistral Large 4 on Hugging Face
- TechCrunchRead ↗Mistral releases weights for its trillion-parameter AI model, targeting cybersecurity
- WiredRead ↗Mistral Says Its AI Model ‘Le Chonk’ Has Leapfrogged China in the Open-Weight Race
- ReutersRead ↗Mistral AI publicly releases Large 4, targets cybersecurity users
- Hacker NewsRead ↗Mistral Large 4 (Le Chonk) is now Open Weight
Details
Mistral released the weights for Mistral Large 4, nicknamed Le Chonk, on October 29, three weeks after its public preview. The multimodal model accepts text and images, has about one trillion parameters in total and 49 billion active parameters according to Hugging Face, and is available through Hugging Face and other platforms. It is distributed under a modified MIT-style license; companies offering it as a commercial cloud service need a separate license from Mistral.
Mistral says additional reinforcement learning improved coding and defensive cybersecurity performance. On Terminal-Bench 4.0, the score rose from 28% in preview to 41%; The Decoder reports that this edges GLM-5.3 at 40%, while Claude Opus 5.5 scored 62% on the Artificial Analysis Coding Agent Index, compared with Large 4 at 48%. The Decoder also places Large 4 third among open models on Artificial Analysis’ Cyber Index at 51%, behind DeepSeek V4.1 Flash at 57% and MiMo-V2.6-Pro at 56%.
The cyber score is reported differently in the supplied sources: Mistral’s announcement and The Decoder give 51%, while Hugging Face’s post says 82% on the Artificial Analysis Cyber Index. Mistral says the model refuses 96.6% of malicious cyber requests; it is also working with select partners on a penetration-testing version that requires review and contractual restrictions. The Decoder reports API prices of $0.70 per million input tokens and $2.10 per million output tokens, and notes that running the model locally requires substantial hardware.
Written from coverage as of