Xiaomi releases MiMo-V2.6 open-weight AI models

Xiaomi has released the MiMo-V2.6 series, a suite of open-weight, omnimodal AI models including the flagship MiMo-V2.6-Pro and the high-efficiency MiMo-V2.6-Flash. The models are designed to process text, images, audio, and video, featuring a 1-million-token context window. According to the Artificial Analysis Intelligence Index, the MiMo-V2.6-Pro has ranked as the top open-weight model globally, tying with Grok 4.7.

The release emphasizes scaled reinforcement learning (RL) training. Xiaomi disclosed that the RL training phase for the Pro model cost approximately $2.62 million and for the Flash model approximately $850,000. The company has also open-sourced over 7,000 RL training environments and its complete training framework under an MIT license to support the research community.

Technical capabilities include 3D spatial reasoning for generating interactive environments and the ability to control robotic hardware. While the models perform at frontier levels, Anthropic has raised concerns, alleging that Xiaomi misused Claude data for large-scale data collection during development.