Details
- Mistral released the improved open weights for Mistral Large 4, nicknamed “Le Chonk,” on October 29, 2026, following an October public preview.[1][2]
- The natively multimodal, 1-trillion-parameter model is available through Mistral’s API and for download from Hugging Face, enabling self-hosted deployment with tools such as vLLM or SGLang.[1][2]
- Three reasoning modes—Low, Medium, and High—let developers trade off performance, latency, and inference cost. Mistral says three additional weeks of reinforcement learning improved coding and defensive cybersecurity performance.[1]
- On Mistral-reported evaluations, Large 4 scored 19.7% on FrontierCode 2.0, 41% on Terminal-Bench 4.0, and 48% on the Artificial Analysis Coding Agent Index while using 47% fewer tokens per task than the cited comparison point.[1]
- The model scored 51% on the Artificial Analysis Cyber Index and reportedly refused 96.6% of malicious cyber requests, while its European training infrastructure and downloadable weights support organizations seeking greater control over AI deployment.[1][2]
Impact
Large 4 strengthens the case for open-weight models in agentic software engineering and defensive security by pairing benchmark gains with lower token usage and customizable deployment. Over the next 12–24 months, its release could increase demand for sovereign, self-hosted AI infrastructure and push model developers to optimize reasoning efficiency, safety controls, and practical coding performance alongside raw benchmark scores.