Mistral Large 4 refreshes open weights after further coding and cyber training
Mistral published a further improved Mistral Large 4 checkpoint after the original preview and weight release, reporting additional reinforcement learning for coding and defensive cybersecurity plus low, medium and high reasoning modes. The open-weight model remains self-hostable under Mistral's ecosystem license.
Why it matters: The new checkpoint makes the promised open model a moving artifact rather than a one-time availability event. Operators can inspect and host it, but need exact checkpoint identity and regression tests because benchmark, refusal and token-efficiency claims…