GPT-6 Astra Distillation: DeepSeek, Moonshot Named in Report
Chinese AI labs DeepSeek and Moonshot AI leaned heavily on GPT-6 Astra distillation to train their newest models, according to a report published by The Information on October 28, citing people with direct knowledge of the matter. The report names Moonshot’s Kimi K3.1 Pro, released on October 23, and DeepSeek’s upgraded V4.1-Flash 1017, launched on October 17.
None of the three companies has confirmed the claim. OpenAI, DeepSeek and Moonshot AI did not immediately respond to requests for comment from Reuters, and no logs, account details or data volumes have been made public.
What GPT-6 Astra Distillation Means
Distillation is a training shortcut. A developer sends a large number of prompts to a stronger model, collects its answers and uses them as training data for its own system. Done at scale, it lets a cheaper model reproduce part of a frontier model’s abilities without the same research and compute budget. Using OpenAI’s outputs this way to build a competing model breaks the company’s terms of service.
Both Chinese models have posted strong scores in their makers’ own tests. Moonshot said Kimi K3.1 Pro beat GPT-6 Astra on a coding benchmark, while DeepSeek has pitched V4.1-Flash as a low-cost alternative to US models. Neither result has been independently verified.
OpenAI Already Pointed at Moonshot in September
The new report builds on an earlier accusation. In a September 30 blog post, OpenAI said it disrupted a coordinated campaign to extract the protected reasoning of its models. The activity started on July 1, spiked on July 24 and 25 with 16,000 requests from more than 4,000 users, and was linked to a wider cluster of more than 15,000 users before OpenAI shut it down by July 28.
OpenAI said it could not tie every operator to a single actor, but it attributed “a core cluster of the activity to individuals associated with Moonshot AI, the developer of Kimi.” The company said the operators did not break its encryption or reach stored user conversations. It banned accounts, closed the replay pathway and shared its findings through the Frontier Model Forum. Moonshot has previously denied that its earlier Kimi K3 model was built through distillation.
Why It Matters
Distillation has become one of the sharpest points of friction in the US and China AI race. Anthropic accused Chinese labs of distilling Claude in September, and on October 12 it said new safeguards had cut the reasoning data harvested by distillers by about 90% to 99%, according to its distillation defenses post.
For OpenAI, the stakes are commercial as well as political. The company is raising money at a reported $1.25 trillion valuation, and its pitch rests on keeping a lead with models such as GPT-6 Astra and the newer GPT-6.1 Terra. If rivals can copy those gains cheaply through API access, that lead narrows fast. For now, the claim against DeepSeek and Moonshot rests on anonymous sources, and we will update this story if any of the companies responds.
Your comment has been submitted,
it will be published after approval.