Xiaomi's MiMo-V2.6-Pro is now the smartest open-weight model

Xiaomi now makes the best open-weight model in the world. MiMo-V2.6-Pro scores 46 on the Artificial Analysis Intelligence Index. That is one point behind GPT-5.6 Sol (max) at 47, and ahead of GLM-5.3 and Kimi K3, which both score 44. Across open and closed models together it ranks sixth. Xiaomi's previous model, MiMo-V2.5-Pro, scored 26.

The bigger story is cost. One Intelligence Index task costs about $0.13 on MiMo, against $1.99 on GPT-5.6 Sol. That puts it on AA's Pareto frontier of intelligence against price. It is a sparse mixture-of-experts model with 1.02T parameters, of which only 42B are active per token. It runs at about 125 tokens/sec, takes text, image, video and audio input and has a 1M-token context window. API pricing is unchanged from V2.5.

Xiaomi released the weights on Hugging Face under the MIT license. The technical report comes with the training environments and RL code. Xiaomi says the reinforcement-learning run took under six days and cost about $2.62M.

Why it matters: an open model you can download, change and ship now comes within a point of the best closed model, for about 7% of the cost.