Mistral Medium 3.5 128B

Mistral Medium 3.5 128B is designed as a unified model for instruction-following, reasoning, and coding rather than a collection of separate specialized systems. Its dense 128B architecture and long context support complex, extended tasks, while reasoning effort can be adjusted per request to balance quick responses with deeper agentic work.

The model also accepts images through a vision encoder trained from scratch for variable sizes and aspect ratios, making it suitable for workflows that combine visual and textual input. It is positioned to replace earlier models in Mistral’s chat and coding-agent products, and a separate EAGLE model is available to accelerate local inference with vLLM or SGLang.