On September 22, in the wee hours of the morning, Xiaomi officially announced and made open-source its latest Xiaomi MiMo-V2.6 series model. This series encompasses two native full-modal models: the Pro and the Flash variants. Owing to the augmentation of RL (Reinforcement Learning) computational capabilities, the MiMo-V2.6-Pro model attained a remarkable score of 46 on the AA Comprehensive Intelligence Index, outperforming Kimi K3 and Qwen3.8 Max, thereby establishing itself as the most formidable open-source model currently available. Nevertheless, when juxtaposed with the leading closed-source models, namely Claude Fable5.1 and GPT-6 Astra, the MiMo-V2.6-Pro still exhibits a discernible performance gap.

The Xiaomi team elucidated that the MiMo-V2.6 series adheres to the API pricing strategy previously employed by the V2.5 series. Notably, the MiMo-V2.6-Pro sets a new benchmark for cost-efficiency among domestic models. At an equivalent level of intelligence, its pricing is merely one-twentieth to one-sixtieth of that charged by overseas counterparts.