Xiaomi's affordable flagship AI leads the open models, and Anthropic says Claude helped get it there
decoder ·
Xiaomi has released its MiMo-V2.6-Pro model, which currently leads the rankings of open AI models while offering significantly lower costs than competitors. The flagship model achieves a score of 46 on Artificial Analysis's Intelligence Index by utilizing a massively expanded reinforcement learning process that took less than six days to complete. To support further development, Xiaomi has openly shared its RL toolkit, including technical reports and thousands of pre-made training tasks for various software and creative domains. However, this release coincides with serious accusations from Anthropic regarding data sourcing practices. Anthropic alleges that Xiaomi engaged in illegal distillation by siphoning approximately 400,000 exchanges through Claude between March and April 2026 to enrich its own training data. This claim suggests that the performance gains credited to reinforcement learning may have been partially derived from improperly using Anthropic's proprietary model outputs. Consequently, Xiaomi stands accused alongside other Chinese labs of abusing Claude to boost their internal model capabilities.
decoder