7d ago2 sources0 reactions

Xiaomi tops open model rankings as Anthropic alleges training data misuse

22 DoomStory toneCapability gain overshadowed by data misuse accusation

By Splitfeed EditorialUpdated Sep 26, 2026, 20:18 UTC

2 sources · Basic Tutorials · The Decoder

Doom: Anthropic accuses Xiaomi of siphoning Claude training data to build MiMo-V2-Pro
Boom: MiMo-V2-Pro reaches top position among openly available AI model benchmarks
Neutral: Reinforcement learning training for the model cost $2.62 million
Boom: Xiaomi's model significantly undercuts competing open models on price

The story in full

Xiaomi released MiMo-V2-Pro, an open AI model that now leads openly available model benchmarks, trained using reinforcement learning at a cost of $2.62 million. The model significantly undercuts competitors on price, according to reporting published September 22, 2026.

Anthropichas accused Xiaomi of siphoning training data from Claude to build the model. The dispute raises questions about the boundaries of acceptable training practices and the use of proprietary model outputs as training signals for competing systems.

Analysis

334 words

By Splitfeed EditorialLast updated Sep 26, 2026, 20:18 UTC

On September 22, 2026, Xiaomi released MiMo-V2.6-Pro, an open AI model that climbed to the top position among openly available models on benchmark rankings. The model was built using large-scale reinforcement learning at a reported cost of $2.62 million, and it significantly undercuts competitors on price. Alongside the release, Anthropic publicly accused Xiaomi of siphoning training data from Claude, the company's proprietary AI system, to develop MiMo-V2.6-Pro.

The accusation matters because it sits at the center of an unresolved debate in AI development: whether using outputs from a proprietary model as training signals for a competing system constitutes misuse, infringement, or something else entirely. Most major AI providers prohibit using their model outputs to train competing models in their terms of service, but enforcement is difficult and the legal boundaries remain largely untested. If Anthropic's claim holds up, it could set a precedent for how the industry handles what is sometimes called model distillation or data siphoning. The competitive dimension is also significant, since a high-performing open model at a fraction of the price of closed alternatives directly pressures the business models of companies like Anthropic, OpenAI and Google.

No reactions from the Pro-AI, Anti-AI or Middle Ground camps have been published yet. Typically, the Pro-AI camp would celebrate a capable open model that democratizes access and drives down costs, while treating the training data dispute as an overreach by proprietary incumbents trying to protect market position. The Anti-AI camp would likely focus on the allegations as evidence of ethical shortcuts and insufficient oversight in how AI companies source training material. The Middle Ground camp would probably call for clearer legal standards around training data use while acknowledging the genuine innovation the model represents.

The dispute will sharpen considerably depending on whether Anthropic files legal action, provides technical evidence of data misuse, or whether Xiaomi responds with a detailed account of its training pipeline. Any regulatory or court filing in the months ahead would give the argument a firmer factual basis to stand on.