🔍 Read the full analysis: The GPT-6 Astra Distillation Report: What It Says About China’s AI Models on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
The Information reported on Oct. 28 that DeepSeek and Moonshot AI relied heavily on outputs from OpenAI’s GPT-6 Astra to train recent models. Reuters reported the allegation, but the companies have not confirmed it, and public details about the scale and method are unavailable.
The Information reported on October 28 that Chinese AI labs DeepSeek and Moonshot AI relied heavily on outputs from OpenAI’s GPT-6 Astra to train recent models, including DeepSeek’s V4.1 Flash update and Moonshot’s Kimi K3.1 Pro. Reuters relayed the report, but the three companies had not confirmed or denied the allegation in its account, and the public evidence does not establish how much Astra output was used or how it was obtained.
The Information’s report, by Jing Yang, Juro Osawa and Qianer Liu, was published on October 28. Its public opening described Astra as OpenAI’s most powerful released model and said Chinese AI labs were among a growing customer base seeking to extract its capabilities. The rest of the article is paywalled. Reuters summarized the central allegation as DeepSeek and Moonshot having “relied heavily” on Astra outputs for their latest models.
The models identified are Moonshot’s Kimi K3.1 Pro, released October 23, and DeepSeek-V4.1-Flash-1017, launched October 17. The source material says Reuters reported that OpenAI, DeepSeek and Moonshot did not immediately respond to its requests for comment. Moonshot’s release post instead credited Kimi’s gains to larger post-training compute and reinforcement learning; DeepSeek’s update note did not describe its training data. Those statements do not independently settle whether Astra outputs were used.
A separate October 29 account by South Korea’s SBS attributed further details to the report, including an alleged shift toward OpenAI after Anthropic tightened defenses. The original article’s full reporting could not be checked against those details because it is paywalled. SBS also quoted an OpenAI spokesperson as saying the company takes the issue seriously and continues working on it. These details remain attributed reporting, not independently established facts.
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
Why Model Training Provenance Matters
If the allegation is substantiated, it could affect how AI companies police access to their models and how business customers evaluate the origins of competing systems. Distillation can lower the cost of building a model that imitates a more capable one, but using a rival’s outputs at scale may breach that provider’s contract. The source material says OpenAI’s terms prohibit using its outputs to build competing models; whether any particular conduct violated those terms would depend on facts not publicly available.
The report also concerns the economics of AI products, though the available figures do not show that the models are equivalent. The source lists Astra at $10 per million input tokens and $50 per million output tokens, Kimi K3.1 Pro at $2 and $10, and DeepSeek’s model at $0.30 and $1.20 during peak hours. It gives scores of 57, 46 and 32, respectively, on the same independent index, but does not identify the index or provide its methodology here. The prices and scores are reported comparisons, not proof that one model can replace another for a company’s specific work.
For organizations choosing models, the immediate practical issue is verification: ask providers about training-data provenance, test models on the tasks that matter, and retain alternatives. The source advises users to avoid cheap, unofficial API resellers, but does not establish that a particular reseller supplied access in this case.
Products Worth Considering
Bose QuietComfort Wireless Bluetooth Headphones with Active Noise Cancellation, Mic, USB‑C Charging, Deep Bass, 24‑Hour Battery – White Smok
LG UltraGear 32GX850A-B 32-Inch 4K UHD Glossy OLED Gaming Monitor
Bose QuietComfort True Wireless Earbuds with Active Noise Cancellation, Black
As an affiliate, we earn on qualifying purchases.
Distillation and Earlier Allegations
Distillation is a standard training technique: a “student” model is trained on responses from a more capable “teacher” model. AI labs use it to create smaller or less expensive systems, including from models they own. The source material quotes a U.S. government advisory describing distillation as a legitimate and useful research technique. The controversy is not the technique alone; it is the reported use of a competitor’s outputs, the scale involved, and whether access was obtained in ways that violate service rules.
The report follows earlier public allegations. The source says OpenAI linked a July campaign to extract hidden reasoning from its models to individuals associated with Moonshot AI and that U.S. security agencies named both DeepSeek and Moonshot in an advisory. It also says China’s commerce ministry called the accusations baseless. These claims and responses form part of the backdrop, but do not verify the new allegation about Astra training data.
The source describes hidden model reasoning as a potential target because detailed reasoning transcripts may convey more than final answers. It cites an Anthropic experiment in which a model trained on full reasoning transcripts from Claude Opus 4.8 scored 80.4% on SWE-bench Verified, compared with 71.8% when trained on summarized transcripts. That is an experiment attributed to Anthropic, not evidence about the training of the DeepSeek or Moonshot models named in this report.
“DeepSeek and Moonshot “relied heavily on outputs” from GPT-6 Astra to train their latest models.”
— The Information, as summarized by Reuters
Evidence Behind the Astra Claim
The public record does not specify how many Astra outputs were used, which prompts or capabilities were involved, or whether the labs accessed the model directly, through resellers, or by another route. No training records or independent technical analysis are provided in the source material. The central claim rests on sources cited by The Information whose identities and evidence are not publicly available in its opening.
There is also no public confirmation or denial from OpenAI, DeepSeek or Moonshot in Reuters’ account. It remains unclear whether either lab disputes the reporting, whether OpenAI has taken enforcement action, and whether the alleged activity would be judged to violate contractual terms or any law. The reported use of outputs should not be treated as proof that the named models were copied wholesale or that their performance came primarily from Astra.
Company Responses and Evidence
The next developments to watch are direct responses from OpenAI, DeepSeek and Moonshot, along with any further reporting that provides details about the alleged scale, method and evidence. OpenAI could also clarify whether it has restricted accounts or pursued action, but the source material does not report a specific enforcement decision.
Until more information is public, companies evaluating these models can ask vendors about data and model provenance, review API terms, and run their own tests. The report does not announce changes to the apps or models, and the source says there is no immediate change for ordinary users. Whether the allegations lead to product, contract or policy changes remains unknown.
Key Questions
Which models does the report name?
It names Moonshot AI’s Kimi K3.1 Pro, released October 23, and DeepSeek-V4.1-Flash-1017, launched October 17.
Has either company confirmed using GPT-6 Astra outputs?
No confirmation is included in the source material. Reuters reported that OpenAI, DeepSeek and Moonshot did not immediately respond to its requests for comment.
Is distillation itself improper?
Not by itself. The source describes distillation as a common technique and quotes a U.S. advisory calling it useful in research. The dispute is about whether rival-model outputs were used at scale and whether access or use breached contractual terms.
Does the report show that the models are as capable as Astra?
No. The source lists different prices and scores on an independent index, but those figures indicate a comparison rather than equivalence. The index methodology is not provided in the material.
What remains unknown?
The public account does not establish the amount of Astra output used, how it was obtained, what evidence supports the claim, or whether OpenAI has taken action. The named companies’ positions on the specific allegation are also not clear from the source material.
Source: ThorstenMeyerAI.com
