Report: DeepSeek and Moonshot trained new models on GPT-6 Astra

What is being claimed?

The Information reported on Wednesday, citing people with direct knowledge of the matter, that Chinese AI companies DeepSeek and Moonshot AI relied heavily on OpenAI's GPT-6 Astra while training their latest models.

The models named in the report are Moonshot's Kimi K3.1 Pro, released last week, and an updated version of DeepSeek's V4.1-Flash, which launched on 17 October.

What distillation means

Distillation is the practice of using the outputs of a stronger "teacher" model to train a smaller or cheaper model. It lets the second model imitate the quality of those outputs at a lower cost than building that capability from scratch.

The technique is not new, but using a rival provider's outputs to train a competing model raises questions about terms of service and competition.

Why it matters

The report highlights how intertwined training methods have become in the US-China model race. If confirmed, it would revive the argument that a US model played a part in the success of leading Chinese ones.

Neither DeepSeek nor Moonshot has commented, and the report rests on a single source and has not been independently verified.