Xiaomi released the weights and training code for its omnimodal MiMo-V2.6 Pro and Flash models on September 21. The Pro version attained a score of 46 on the Artificial Analysis Intelligence Index, the highest for any open-source model, matching the performance of Claude Opus 5 and GPT-5.6 Sol on agent benchmarks. A reinforcement learning scaling run to develop the series cost $2.6 million for Pro and $0.9 million for Flash.

The Pro model features a 1 million token context window and costs $0.435 per 1 million input tokens and $0.87 per 1 million output tokens. Xiaomi researchers estimated these models provide a 10x productivity increase in materials science R&D, shortening design cycles from one month to three days. The release also includes a proprietary end-to-end reinforcement learning framework and 7,000 task environments.