DeepSeek and Moonshot AI leaned heavily on outputs from OpenAI’s GPT-6 Astra to train their newest models, Kimi K3.1 Pro and an October 17 upgrade to DeepSeek V4.1-Flash, according to The Information. Copying a rival model’s outputs this way, known as distillation, mostly targeted Anthropic’s Claude until Anthropic said this month that new defenses had cut the data distillers could pull from it by 90% to 99%, so Chinese labs shifted to Astra. OpenAI said it’s banning fraudulent accounts and tightening signup checks to stop the practice, but people familiar with the Chinese labs’ work said distillers get around those checks by spreading requests across many accounts.
