Overview
- Mistral publicly posted the weights for Mistral Large 4, nicknamed Le Chonk, on Hugging Face and opened API access on Thursday, October 29, 2026.
- The company describes the model as a roughly 1 trillion‑parameter mixture‑of‑experts system with a 1 million‑token context window that can run on as few as four Nvidia B200/B300 GPUs.
- Mistral says reinforcement‑learning fine‑tuning since its Oct. 6 preview lifted a Terminal‑Bench 4.0 score from about 28% to 41%, which the firm cites as a narrow lead among open‑weight models for agentic coding tasks.
- The firm markets ML4 for cybersecurity work, reporting high refusal rates on malicious prompts and strong CyberGym E2E scores, but independent benchmarking and safety audits of those claims are still pending.
- Mistral positions the open‑weights release as a European, sovereign option that lowers dependency on closed APIs, even as closed proprietary models from U.S. labs continue to outperform on broad capability benchmarks.