Xiaomi MiMo-V2.6-Pro tops open models on Artificial Analysis index as Anthropic distillation claims resurface
Xiaomi has released open-weight omnimodal models MiMo-V2.6-Pro and MiMo-V2.6-Flash, and reporting from The Decoder places Pro at the top of openly available models on the Artificial Analysis Intelligence Index with a 46-point score while advertising sub-dollar per-million-token API pricing. The same coverage ties that lead to a reinforcement-learning investment priced at about $2.62 million for Pro. Techmeme carries Xiaomi's claim that Pro performs on par with frontier closed models across most agent benchmarks. The breakthrough narrative is tempered by Anthropic naming Xiaomi in threat intelligence over alleged siphoning of Claude training data, an accusation the packet presents without corroborating evidence or a Xiaomi response. Builders weighing adoption face a familiar tension between attractive open-model economics and unresolved data-provenance risk.
Xiaomi MiMo-V2.6-Pro tops open models on Artificial Analysis index as Anthropic distillation claims resurface
The Decoder reports Xiaomi shipped MiMo-V2.6-Pro and MiMo-V2.6-Flash after a large reinforcement-learning run it prices at about $2.62 million for Pro. The piece cites a 46-point Artificial Analysis Intelligence Index score and sub-dollar per-million token API pricing, while recounting Anthropic naming Xiaomi in a threat-intelligence case over alleged Claude training-data siphoning.
Key takeaway
Treat MiMo-V2.6-Pro's benchmark and pricing story as reported, not verified, until independent checks land and Anthropic's data-siphoning allegation is adjudicated.
What happened
The Decoder reports that Xiaomi shipped MiMo-V2.6-Pro and MiMo-V2.6-Flash after a large reinforcement-learning run it prices at about $2.62 million for Pro, citing a 46-point Artificial Analysis Intelligence Index score and sub-dollar per-million token API pricing for the flagship open release.
The same reporting recounts Anthropic naming Xiaomi in a threat-intelligence case over alleged Claude training-data siphoning, while Techmeme notes Xiaomi's public debut of the open-weight omnimodal series and its claim that Pro performs on par with Opus 5 and GPT-5.6 Sol across most agent benchmarks.
Evidence
The Decoder ties MiMo-V2.6-Pro to a 46-point Artificial Analysis Intelligence Index score and sub-dollar per-million token API pricing.
The Decoder · attributed
The Decoder reports Xiaomi shipped MiMo-V2.6-Pro and MiMo-V2.6-Flash after a large reinforcement-learning run it prices at about $2.62 million for Pro. The piece cites a 46-point Artificial Analysis Intelligence Index score and sub-dollar per-million token API pricing
The Decoder reports reinforcement learning for Pro priced at about $2.62 million.
The Decoder · attributed
What makes that possible is massive reinforcement learning that cost $2.62 million.
Anthropic accuses Xiaomi of siphoning training data from Claude, per The Decoder.
The Decoder · attributed
Anthropic accuses the company of siphoning off training data from Claude.
Techmeme reports Xiaomi open-sourced MiMo-V2.6 Pro and Flash with a frontier agent-benchmark claim.
Techmeme · attributed
Today, we are releasing and open-sourcing the MiMo-V2.6 series.
Rhodium research cited via CNBC puts OpenAI and Anthropic revenue at ten times all Chinese AI models combined.
CNBC AI · attributed
OpenAI and Anthropic are making 10 times more revenue than all Chinese AI models combined, research group Rhodium says CNBC
The New Stack reports open-weight models account for a majority of tokens on Vercel's AI Gateway while Anthropic still takes 64% of spend.
The New Stack AI · attributed
Open-weight models now handle a majority of tokens on Vercel’s AI Gateway. But Anthropic still takes 64% of the spend.
Why it matters
If verified, a top-ranked open model at aggressive API pricing would sharpen competition for builders seeking alternatives to closed frontier APIs, while an unresolved Anthropic allegation would raise compliance and reputational stakes for production adoption.
Limits and uncertainties
The Decoder excerpt is truncated and presents performance, pricing, and the $2.62 million RL figure as reported claims rather than independently verified measurements.
Anthropic's training-data siphoning allegation is described as an accusation without supporting evidence in the packet, and no Xiaomi response appears in the corpus.
Practical implications
Operators evaluating MiMo-V2.6-Pro should validate Artificial Analysis scores, live API pricing, and license terms before routing production traffic.
Legal and security teams should track Anthropic threat-intelligence disclosures and any formal enforcement before treating the weights as clean-room open source.
What to watch
Independent replication of the 46-point Artificial Analysis Intelligence Index result and published MiMo-V2.6 API price sheets.
Any public Xiaomi rebuttal or documented legal action tied to Anthropic's Claude training-data siphoning allegation.
