OpenAI’s Astra Model Becomes Top Distillation Target for China’s AI Labs

Jing Yang, Juro Osawa and Qianer Liu report that DeepSeek and Moonshot AI relied heavily on outputs from OpenAI's GPT-6 Astra to train their latest models (Moonshot's Kimi K3.1 Pro and DeepSeek's V4.1-Flash 1017 upgrade), per people with direct knowledge. Until recently Anthropic's Claude was the main target; Anthropic says new defences cut data harvested by distillers by about 90-99%, and its IPO prospectus says customers violating its terms contributed roughly 5-7% of revenue in the first nine months (revenue through September over $33B). OpenAI declined to estimate the revenue effect but says it takes the issue 'extremely seriously'; its annualised revenue neared $50B last month. Caveats: Moonshot didn't comment and DeepSeek didn't respond, Chinese labs haven't publicly answered the allegations, and the article notes distillation isn't the only reason Chinese models are competitive.