Xiaomi released weights for its MiMo-V2.6 Pro and Flash omnimodal models under an MIT license. The Pro version is the top open-weight model on the Artificial Analysis Intelligence Index with a score of 46, surpassing competitors Grok 4.7 and Kimi K3. On the Code Arena WebDev benchmark, the model earned 1,628 points, which ties Claude Fable 5 and represents a 153 point gain over the previous MiMo-V2.5-Pro.

The Pro model uses a Mixture of Experts architecture with 1.02T total and 42B active parameters, while the Flash version has 309B total and 15B active. Alongside the models, Xiaomi provided an end-to-end RL training framework and 7,000 diverse RL environments to assist the open-source community. The Flash variant supports a 1M token context window and reaches a decode speed of 250 tokens per second.