🔍 Read the full analysis: Did China’s New AI Models Learn From GPT-6 Astra? What We Can And Can’t Say on ThorstenMeyerAI.com
Get the little things that make your day 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 Oct. 28 that DeepSeek and Moonshot AI relied heavily on GPT-6 Astra outputs to train recently released models. Reuters repeated the claim, but the public record does not provide methods or figures, and the companies had not responded to Reuters. Distillation is a standard AI technique; whether rivals’ outputs were obtained and used in a way that breaches terms remains unconfirmed.
The Information reported on Oct. 28 that China-based AI labs DeepSeek and Moonshot AI relied heavily on outputs from OpenAI’s GPT-6 Astra to train recent models. Reuters repeated the report within hours, but the public account relies on unnamed sources, and none of the three companies had responded to Reuters when it published its report.
The models identified were Moonshot’s Kimi K3.1 Pro, released Oct. 23, and DeepSeek-V4.1-Flash-1017, released Oct. 17. Reuters reported that the models “benefitted from distilling Astra,” attributing the account to The Information. The original article’s public opening described Chinese AI labs as a growing customer base for Astra; the rest of its reporting is behind a paywall.
The available material does not say how many outputs were used, how they were acquired or what portion of either model’s training depended on them. Moonshot’s release post attributed Kimi K3.1 Pro’s improvements to substantially larger post-training compute and large-scale reinforcement learning. DeepSeek’s update note did not discuss its training data. Those statements do not independently confirm or disprove The Information’s account.
A Korean broadcaster, SBS, summarized additional details on Oct. 29, including that labs had shifted toward OpenAI after Anthropic tightened defenses. SBS also quoted an OpenAI spokesperson as saying the company takes the issue seriously. Those details have not been checked against the full paywalled report in the source material, so they remain separately attributed to SBS.
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 Provenance Matters
If the report is accurate, it raises questions about how competing AI systems are trained and whether model providers can protect outputs and restricted reasoning from large-scale collection. For businesses choosing a model, the story also makes training-data provenance and access practices relevant parts of vendor review, alongside price and performance.
It does not establish that either company broke the law. OpenAI’s terms prohibit using its outputs to build competing models, according to the source material. Whether the reported activity occurred, how access was obtained and whether the terms apply to specific conduct would matter to any contractual dispute. Possible deception or account misuse is not established by the public reporting.
The commercial stakes are tangible but should not be confused with proof of copying. The source gives Astra prices of $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, their scores were 57, 46 and 32, respectively. The index and prices provide a snapshot, not evidence about training or a guarantee of performance on a company’s own work.
Products Worth Considering
LEVOIT Vital 200S-P Smart Air Purifier for Large Rooms
Levoit Core 400S-P Smart Air Purifier for Large Rooms
Oral-B iO Series 9 Limited Rechargeable Electric Toothbrush with 7 Cleaning Modes
As an affiliate, we earn on qualifying purchases.
Distillation and Earlier Allegations
Distillation is a common way to train a smaller model using answers generated by a larger “teacher” model. AI labs use the method on their own systems to produce less costly models. A U.S. government advisory cited in the source material describes distillation as “a legitimate and useful technique in AI research.” The dispute here is not the technique by itself, but the reported use of a competitor’s outputs at scale and the circumstances under which those outputs may have been obtained.
The story follows earlier accusations. The source says that in September, OpenAI linked a July campaign to extract hidden reasoning from its models to people associated with Moonshot AI. It also says U.S. security agencies named DeepSeek and Moonshot in an advisory, while China’s commerce ministry called the accusations baseless. These are claims and responses from the parties and agencies involved; they do not independently verify the new report about Astra.
Separately, the source cites a February 2026 Anthropic account of more than 16 million exchanges with Claude involving DeepSeek, Moonshot and MiniMax. The supplied excerpt ends before giving further details, so it cannot establish a direct connection to the present allegation.
“DeepSeek and Moonshot “relied heavily on outputs from OpenAI’s powerful GPT-6 Astra model to train their latest models.””
— The Information, as summarized by Reuters
Evidence Behind the Astra Claim
The claim remains unverified publicly. The Information’s report, as available in the source material, cites unnamed people with direct knowledge; its detailed reporting is paywalled. No data is given on the volume of Astra outputs, the prompts used, the accounts involved or the role those outputs played in either model’s training.
Reuters said OpenAI, DeepSeek and Moonshot did not immediately respond. That is not a confirmation or denial, and the source material does not provide later responses. It also does not establish whether any terms were breached, whether access involved fake or resold accounts, or whether the reported activity would amount to unlawful conduct.
Company Responses and Verification
The next useful developments would be responses from OpenAI, DeepSeek and Moonshot, alongside further reporting that clarifies the sources, evidence and scale behind the allegation. Any account of account access, collection methods or use in training would help separate the reported claim from what can be independently confirmed.
For now, companies evaluating models can ask providers about training-data provenance, keep a tested alternative and assess models on their own tasks. Users should distinguish official services from unofficial API resellers. The report does not describe a change to the models or their availability, and the source material says nothing changes in the apps immediately.
Key Questions
What did The Information report?
It reported that DeepSeek and Moonshot AI relied heavily on GPT-6 Astra outputs to train recent models, naming Kimi K3.1 Pro and DeepSeek-V4.1-Flash-1017. Reuters repeated the claim; the public record described here does not include its underlying data.
Has either company confirmed the report?
No confirmation is provided in the source material. Reuters said OpenAI, DeepSeek and Moonshot did not immediately respond to its requests. That does not establish their later positions.
Is AI distillation itself improper?
Not inherently. Distillation is a standard training technique, including for labs’ own models. The disputed issues are whether a company used a rival’s outputs, at what scale, and whether the method of access or use violated applicable terms or law.
Does the report prove a legal violation?
No. The report does not publicly establish how outputs were obtained or whether conduct breached OpenAI’s terms. A contractual or legal conclusion would require more evidence and analysis.
What should businesses and users do now?
Businesses can ask vendors about data provenance and test models on their own workloads. Users should use official services and be cautious with unofficial API resellers. The source material reports no immediate change to the apps.
Source: ThorstenMeyerAI.com
