🔍 Read the full analysis: Did China’s Newest AI Models Draw On GPT-6 Astra? A Report Review on ThorstenMeyerAI.com
Get the latest gadgets delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
TL;DR
The Information reported on October 28 that DeepSeek and Moonshot AI relied heavily on OpenAI’s GPT-6 Astra outputs to train recent models. Reuters reported that none of the three companies immediately responded; the public reporting does not detail the methods or scale, and the claim remains unconfirmed by the companies.
The Information reported October 28 that Chinese AI labs DeepSeek and Moonshot AI relied heavily on outputs from OpenAI’s GPT-6 Astra to train recent models, a claim that could raise questions about access to a rival’s systems and the terms governing their use. Reuters summarized the report, but the companies have not publicly confirmed it, and the available reporting does not explain how many outputs were involved or how they were obtained.
The Information, in a report by Jing Yang, Juro Osawa and Qianer Liu, named Moonshot’s Kimi K3.1 Pro and DeepSeek-V4.1-Flash-1017. According to Reuters’ account, the first was released October 23 and the DeepSeek update on October 17. The Information said the models “benefitted from distilling Astra”; Reuters described the claim as the labs having relied heavily on Astra outputs. The original story’s full text is behind a paywall, limiting what can be independently checked from its public opening.
Reuters reported that OpenAI, DeepSeek and Moonshot did not immediately respond to requests for comment. A separate October 29 summary by South Korean broadcaster SBS attributed to an OpenAI spokesperson a statement that the company takes the issue seriously and is continuing to work on it. The statement and additional details in SBS’s account could not be checked against the paywalled report in the source material.
The public record does not provide training-data counts, collection methods, account details or technical evidence showing how Astra outputs were used. Moonshot’s Kimi launch post attributed its model’s gains to substantially larger post-training compute and large-scale reinforcement learning. DeepSeek’s update note, as described in the source material, did not discuss its training data. Those company materials do not settle the allegation either way.
Infographic 01 · Distillation in plain words
How distillation works, and where it crosses a line.
A big model teaches a small one by example. The technique is ordinary; the dispute is about whose model is the teacher and how the answers were obtained.
Send the teacher prompts
A strong “teacher” model gets a huge number of carefully chosen tasks.
Keep answers and reasoning
The step-by-step working is worth far more than the final answer alone.
Teach a student to imitate
A smaller or cheaper “student” model learns to reproduce the teacher’s moves.
Normal practice
The U.S. advisory calls it “a legitimate and useful technique”.
- A lab distills its own large model
- Goal: smaller, faster, cheaper versions
- Open models used within their licence
Where the fight starts
What OpenAI, Anthropic and U.S. agencies allege.
- The teacher is a rival’s model
- Industrial scale: millions of exchanges
- Fake, resold or stolen accounts and keys
- Breaks terms that ban building competing models
Why hidden reasoning is the prize
Anthropic’s own test: a model trained on Claude Opus 4.8 transcripts, scored on SWE-bench Verified (coding).
That is why providers now hide or summarize reasoning, and why extraction attempts target it.
Distillation is a student copying a top tutor’s worked solutions. Legal in a classroom. Contested when the tutor never agreed.Thorsten Meyer · my reading
Sources: NSA/CISA/FBI advisory AA26-251A, 8 Sep 2026; Anthropic, 12 Oct 2026 (vendor’s own experiment). Analogy and framing: my reading.
Powered by Thorsten Meyer AI
The Dispute Over Model Outputs
Distillation is a standard training technique: a smaller model learns to reproduce responses generated by a more capable “teacher” model. Labs commonly apply it to their own systems. The central dispute here is not the technique by itself, but whether a company used a competitor’s outputs at scale and whether access complied with that provider’s rules.
OpenAI’s terms prohibit using its outputs to build competing models, according to the source material. If the reported conduct occurred, it could prompt a contract dispute; claims that access was obtained through fake or resold accounts could raise separate questions about how those accounts were used. The available report does not establish that any particular access method was used, and whether the terms were breached would depend on evidence and legal interpretation.
The allegation also matters to businesses choosing AI systems. The source lists Astra at $10 per million input tokens and $50 per million output tokens, compared with $2 and $10 for Kimi K3.1 Pro and $0.30 and $1.20 for DeepSeek’s model at peak hours. On the same independent index cited there, the models scored 57, 46 and 32 respectively. These prices and scores are reported figures, not proof of how any model was trained; they suggest lower listed costs do not mean equal scores on that index.
Popular right now
TCL QM64L 85-Inch QD-Mini LED 4K Smart Fire TV
Vitamix Propel Series 750 Professional-Grade Blender, 4 Programs, Black
Sony BRAVIA 5 85-Inch Mini LED 4K HDR Smart TV (K-85XR50)
As an affiliate, we earn on qualifying purchases.
Earlier Concerns Over Distillation
The allegation follows earlier public scrutiny of how AI labs obtain competitors’ model outputs. In September, OpenAI linked what it called a “core cluster” of a July campaign to extract hidden reasoning from its models to individuals associated with Moonshot AI, according to the source material. U.S. security agencies also named DeepSeek and Moonshot in an advisory. These are prior claims and government statements, not independent proof of the conduct described in the latest report.
China’s commerce ministry called the accusations baseless, according to the source. The competing accounts underline why attribution matters: the new report relies on unnamed people described as having direct knowledge, while the companies have not publicly confirmed its central claim. A U.S. government advisory cited in the source describes distillation as a “legitimate and useful technique in AI research,” while warning, in the broader dispute, about how it may be used.
The source also says SBS reported that Chinese labs had mainly copied Anthropic’s Claude until Anthropic tightened defenses, then shifted toward OpenAI. That detail is not independently verifiable from the original paywalled article based on the material available. The public launch posts from Moonshot and DeepSeek provide their own descriptions of training, but neither discloses enough information to verify or disprove the reported use of Astra.
Evidence and Methods Still Missing
The central allegation remains unconfirmed by OpenAI, DeepSeek and Moonshot in the reporting available here. The underlying report relies on unnamed sources, and its full text is paywalled. No publicly described dataset, prompt logs, account records or independent technical analysis establishes the scale or method of any alleged distillation.
It is also unclear whether the models’ training relied heavily on Astra outputs, what proportion of their capabilities may have come from such outputs, or whether any access violated OpenAI’s terms. The labs’ launch descriptions give alternative accounts of training choices, but they do not provide enough detail to resolve those questions. The separate SBS account adds claims about earlier use of Anthropic’s Claude and an OpenAI response, but those details are not independently verified in the source material.
Responses and Evidence to Watch
The next developments to watch are formal responses from OpenAI, DeepSeek and Moonshot, along with any additional reporting that discloses the evidence behind the allegation. Technical documentation, account-access records or other verifiable material could clarify what outputs were used and at what scale. Until then, the report should be treated as an attributed allegation rather than an established finding.
For companies evaluating models, the source advises checking model provenance, keeping a fallback provider and testing systems on their own tasks. Users do not have a confirmed reason, based on this report alone, to expect changes in the named products’ apps. The source cautions users against unofficial, unusually cheap API resellers; the report itself does not establish that any particular reseller was involved.
Key Questions
Which models are named in the report?
The report names Moonshot’s Kimi K3.1 Pro and DeepSeek-V4.1-Flash-1017 as models that allegedly benefited from distilling GPT-6 Astra outputs.
Have the companies confirmed the allegation?
No. Reuters reported that OpenAI, DeepSeek and Moonshot did not immediately respond to requests for comment. The allegation has not been publicly confirmed by those companies in the available reporting.
Is distillation itself improper?
No. Distillation is a recognized training technique and is used by labs to train smaller models. The dispute concerns whether a rival’s outputs were used at scale and whether that use complied with the provider’s terms.
What evidence has been made public?
The available source material describes an unnamed-source report and Reuters’ summary, but gives no output counts, access method or public technical evidence. The full Information article is paywalled, and the companies have not confirmed the claim.
Will this change what users see in the apps?
The report does not establish any immediate change to the named models or their apps. Further company responses or evidence could clarify whether there are product, access or policy consequences.
Source: ThorstenMeyerAI.com
