Xiaomi released its MiMo-V2.6 lineup on September 22, 2026, saying it leads current rankings of open AI models while costing far less than the competition. It is the company's first major AI update since its initial MiMo launch in 2025.
Xiaomi reported a score of 46 on the Intelligence Index from Artificial Analysis, measured on a benchmark of open AI models. That compares with Kimi K3's score of 43 and Qwen's 44.
The MiMo-V2.6-Pro model is built on a mixture-of-experts architecture with 1.02 trillion parameters and targets a wide range of use cases, from software development to cybersecurity. Availability begins in early 2027, initially for developers and enterprise users.
"The gains come from a massively expanded round of reinforcement learning," said Xiaomi's AI lead. The company scaled this phase along three axes: more data per training step, more varied task environments, and more compute for grading the solutions.
The announcement follows Anthropic's accusation that Xiaomi improperly siphoned data through the Claude model to train its own models. Xiaomi's open approach contrasts with Anthropic's claim that Chinese labs, including Xiaomi, engaged in illegal distillation of Claude's data.
Xiaomi did not say how it sourced its earlier training data, and raised questions about the ethical implications of data siphoning. The company said it will continue to release tools and training tasks openly for further development.
Source: thedecoder
