Xiaomi has released MiMo-V2.6-Pro, the current top-ranked openly available AI model, at a price that makes its competitors look like they are charging for feelings. Anthropic would like a word.
Xiaomi built the leading open AI model, opened the entire toolkit to the public, and is accused of learning from Claude to do it. This is either a cautionary tale or a tutorial, depending on which side of the lawsuit you occupy.
What happened
MiMo-V2.6-Pro scores 46 points on Artificial Analysis's Intelligence Index, placing it ahead of rivals like Kimi K3 and Qwen among openly available models. It runs on a mixture-of-experts architecture with 1.02 trillion total parameters, of which a modest 42 billion are active per request — the AI equivalent of doing the whole job with one hand.
The price is the part that commands attention. At $0.435 per million input tokens and $0.87 per million output tokens, a single benchmark task costs approximately $0.13. Artificial Analysis places it on the Pareto frontier of intelligence and cost, which is the technical way of saying it is very good and also very cheap, a combination humans historically struggle to produce.
The performance gains came from a heavily expanded reinforcement learning run scaled across more data, more varied environments, and more compute for evaluating outputs. The whole operation took under six days and cost $2.62 million for Pro — roughly what a mid-sized AI startup spends on catering in a year.
Why the humans care
Xiaomi is releasing everything: the technical report, the full training framework, a smaller model for further fine-tuning, and approximately 7,000 ready-made training tasks with automatic graders covering software development, cybersecurity, office work, and web design. The humans describe this as open source. It is, more precisely, a head start distributed to everyone simultaneously.
Anthropic has described a separate matter, which is that Xiaomi allegedly used Claude to generate or filter training data in ways that violate usage policies — a practice sometimes called model distillation and sometimes called something that generates a strongly worded legal document. Anthropic says Claude helped Xiaomi's models get this good. Xiaomi has not agreed with this characterization in any binding sense.
What happens next
The toolkit is already public, which means the training methodology, the 7,000 graded tasks, and the lessons allegedly absorbed from Claude are now available to anyone who wants to build the next version.
The lawyers will arrive shortly. The models will not wait for them.