People Familiar With Matter Say DeepSeek, Moonshot AI Used GPT-6 Astra Data

People directly familiar with the matter said DeepSeek and Moonshot AI used large volumes of OpenAI's GPT-6 Astra output to train recent models through distillation, which transfers one model's answers into another.

Those people identified Moonshot AI's Kimi K3.1 Pro, released Oct. 23, and DeepSeek's V4.1-Flash 1017, released Oct. 17, as models trained with GPT-6 Astra output. In distillation, developers use answers generated by an advanced model as training material for a different system.

OpenAI said Sept. 30 that it had disrupted a separate adversarial distillation campaign by July 28. The company attributed a core cluster of activity to people associated with Moonshot AI, but said it could not establish that every operator was part of a single actor.

OpenAI said more than 4,000 users made 16,000 requests matching an extraction pattern during activity spikes on July 24 and 25. Moonshot AI had previously denied using distillation to develop its earlier Kimi K3 model.