Xiaomi officially announced and open-sourced the brand-new Xiaomi MiMo-V2.6 series of all-modal models early this morning, comprising two distinct versions: Pro and Flash. This release signifies a substantial leap in Xiaomi's pursuit of the RSI pathway, showcasing a remarkable boost in model intelligence through an expansion of RL computing capabilities. In the AA Comprehensive Intelligence Index v4.3.2, the MiMo-V2.6-Pro achieved a score of 46 points, securing its position as the top-ranked open-source weight model under this benchmark. Although there remains a noticeable performance gap when compared to elite closed-source models, its capabilities have significantly advanced from the previous generation. The MiMo-V2.6 series stands out among domestic open-source models for its relatively high investment in RL computing power. Both versions were trained within a span of less than six days, incurring costs of approximately $2.62 million and $850,000, respectively. The series has demonstrated enhancements in training task pass rates and long-term software engineering evaluation scores. The series enhances RL computing power across three key dimensions: increased batch sizes and throughput, a broader range of tasks and complex environments, and enhanced Grader computing capabilities. The models incorporate 3D spatial reasoning, multimodal perception, and computer operation skills, making them suitable for applications such as game development and embodied simulation. They can also aid researchers in material design screening and the formal proof of mathematical theorems. Regarding pricing, the MiMo-V2.6 series adheres to the API pricing strategy of its forerunner and introduces the MiMo Desktop client along with a membership subscription plan, with the client featuring an UltraSpeed mode. Furthermore, Xiaomi has completely open-sourced the model weights, technical reports, and accompanying RL research resources of this series. This includes over 7,000 high-quality RL task environments, an end-to-end RL training framework, and a lightweight, modular Harness, all designed to foster research and innovation within the community.