🔍 Read the full analysis: China’s New AI Models: Reading Between The Lines Of The GPT-6 Astra Report on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
The Information reported that DeepSeek and Moonshot AI relied heavily on outputs from OpenAI’s GPT-6 Astra when developing recent models. Reuters reported that none of the three companies responded to its requests for comment. The public record does not disclose the scale or method of the alleged use, and the claim remains unconfirmed by the companies.
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 recently released models. Reuters summarized the report, but said OpenAI, DeepSeek and Moonshot did not immediately respond to requests for comment; the public account does not establish how much Astra output was used or how it was obtained.
The models named in the reporting are Moonshot’s Kimi K3.1 Pro, released October 23, and DeepSeek-V4.1-Flash-1017, released October 17. Reuters said The Information reported that both models “benefitted from distilling Astra.” The Information’s article, by Jing Yang, Juro Osawa and Qianer Liu, appeared October 28 at 6 p.m. Pacific time. 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.
The allegation currently rests on unnamed sources cited by The Information, as relayed by Reuters. The source material provides no figures for the volume of Astra responses involved, no description of the training process, and no evidence about whether access came through ordinary accounts, resellers or deceptive means. Reuters said the three companies did not immediately comment. That lack of response is not confirmation of the allegation.
Moonshot’s launch post attributed Kimi K3.1 Pro’s gains to substantially larger post-training compute and large-scale reinforcement learning. DeepSeek’s update note did not disclose training data, according to the source material. Those statements do not independently resolve whether Astra outputs were also used.
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 Stakes of Model Distillation
Distillation is a common training method: a smaller model learns to reproduce answers from a larger “teacher” model. The U.S. government advisory cited in the source material describes the technique as “a legitimate and useful technique in AI research.” The disputed issue is not the method by itself, but the reported use of a rival’s model at scale and whether the access and use complied with the provider’s terms.
The source material says OpenAI’s terms prohibit using its outputs to build competing models. If that applies to the conduct alleged, the dispute could center on contract terms and how access was secured; any legal conclusion would depend on facts not publicly available. The report does not establish that the labs used fake accounts, resold access or otherwise circumvented controls.
For customers, the practical issue is both cost and provenance. 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 one independent index cited by the source, the models scored 57, 46 and 32, respectively. Those figures suggest lower listed prices alongside lower scores on that index, not equivalent performance; the index’s name and testing details are not provided here.
Popular right now
De'Longhi Magnifica Evo Fully Automatic Espresso Machine with Manual Milk Frother
De'Longhi Rivelia Fully Automatic Espresso Machine with Milk Frother, White
Roborock Qrevo CurvX Robot Vacuum and Mop
As an affiliate, we earn on qualifying purchases.
Earlier Scrutiny of AI Outputs
The report arrives amid previous disputes over the use of AI systems’ outputs. According to the source material, OpenAI said in September that it linked a “core cluster” of a July campaign to extract hidden reasoning from its models to individuals associated with Moonshot AI. The source also says U.S. security agencies named DeepSeek and Moonshot in an advisory, while China’s commerce ministry called the accusations baseless. These are separate claims and responses; they do not independently prove the new allegation about Astra.
The source also recounts an earlier Anthropic account from February 2026 involving DeepSeek, Moonshot and MiniMax. It says Anthropic reported more than 16 million exchanges with Claude, but the supplied material cuts off before giving further details. A Korean broadcaster, SBS, reportedly described claims that Chinese labs had previously focused on Anthropic’s Claude and shifted toward OpenAI after Anthropic tightened defenses. The article’s author said those details could not be checked against The Information’s paywalled report, so they remain secondhand and unverified in the material provided.
“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
What the Public Record Lacks
The central claim remains unconfirmed by OpenAI, DeepSeek and Moonshot in the reporting provided. The available account does not specify the number of Astra outputs used, the prompts involved, which parts of the models’ training were affected, or whether the alleged use materially changed either model’s capabilities.
It is also unclear whether access involved ordinary paid use, shared or resold accounts, fake identities, or another method. The distinction matters because the source material describes possible violations of OpenAI’s terms and possible access-related fraud, but does not establish that either occurred. The paywalled article’s evidence cannot be fully assessed from its public summary, and there is no independent technical analysis in the supplied material demonstrating that the released models reproduce Astra outputs.
Responses and Evidence to Watch
The next useful developments would be on-the-record responses from OpenAI, DeepSeek and Moonshot, along with more detail from The Information about the sources, scale and methods behind its reporting. Technical evidence, account records or a formal statement from a regulator could help clarify whether the alleged training occurred and whether it breached contracts or other rules.
For companies choosing models, the source recommends checking data provenance, testing systems on their own workloads and maintaining a fallback option. Users have no reported app changes tied to this story. The source advises caution with cheap, unofficial API resellers, although it does not report that any particular reseller was involved in the alleged conduct.
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. The claim is based on unnamed sources in the account summarized by Reuters.
Have the companies confirmed the report?
No. Reuters said OpenAI, DeepSeek and Moonshot did not immediately respond to its requests for comment. The supplied material includes no confirmation from any of them.
Is distillation itself improper?
Not inherently. Distillation is a standard way to train a smaller model using a larger model’s outputs. The questions raised here concern whether a competitor’s outputs were used at scale and whether that use complied with access rules and contractual terms.
What is still unknown about the allegation?
The public account does not state how many outputs were used, how they were obtained, or how much they influenced either model. It also does not provide independent technical evidence confirming the reported training method.
Does the report change what users should do now?
The source reports no immediate change to the apps. It advises users to avoid cheap, unofficial API resellers; the report does not identify a specific reseller as part of the allegation.
Source: ThorstenMeyerAI.com
