.png){kind=logo}
Chinese AI developers DeepSeek and Moonshot AI relied heavily on outputs from OpenAI’s GPT-6 Astra to train their latest models, according to a report by The Information. Reuters relayed the report, citing people with direct knowledge of the matter.
The claim reignited a debate over distillation, a technique in which developers use the outputs of a stronger model to train another model. Leading AI firms argue that the practice helps rivals reproduce advanced capabilities quickly and at a lower cost.
The report names two recent releases: Moonshot AI’s Kimi K3.1 Pro, launched last week, and DeepSeek’s upgraded V4.1-Flash 1017, released on October 17. Experts suggest that these benefited from distilling GPT-6 Astra.
The report does not specify how the companies accessed Astra outputs, how much data they used, or whether they violated any contractual or platform rules. It also does not establish that the companies broke the law.
OpenAI, DeepSeek and Moonshot AI did not immediately respond to Reuters’ requests for comment. However, the companies have not confirmed or denied the reported use of Astra outputs.
Distillation has become a major flashpoint as competition between US and Chinese AI developers intensifies. OpenAI, Anthropic and Google have argued that it allows competitors to replicate frontier capabilities without bearing the full cost of original research.
Anthropic said new safeguards against model-copying had reduced data harvested in so-called distillation attacks by 90–99 percent. The development underscores how leading labs are tightening access to their models.
The allegations arrive at a sensitive moment for China’s AI sector. DeepSeek and Moonshot are among China’s most prominent model developers, and their latest releases have been presented as challengers to leading Western AI models.
Also Read: Gemini is Becoming More Than a Chatbot as Google Opens the Door to More Apps
