🔍 Read the full analysis: China’s Latest AI Models And GPT-6 Astra: What The Report Can Tell Us on ThorstenMeyerAI.com
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
The Information reported on October 28 that DeepSeek and Moonshot AI relied heavily on outputs from OpenAI’s GPT-6 Astra when developing recent models. Reuters summarized the report, but the companies have not confirmed or denied the claim, and public details about the evidence, scale and methods are limited.
The Information reported October 28 that China’s 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 technology and compliance with OpenAI’s terms. Reuters relayed the report, but none of the three companies had publicly confirmed or denied it in Reuters’ account.
The models named are Moonshot’s Kimi K3.1 Pro, released October 23, and DeepSeek-V4.1-Flash-1017, released October 17. The Information’s public opening says Astra has attracted demand from Chinese AI labs seeking to extract its capabilities; the rest of its report is behind a paywall. Reuters summarized the central allegation as the labs having relied heavily on Astra outputs.
The public account does not provide figures or a method for how many outputs were used, how they were obtained, or what share of either model’s training came from them. Reuters said OpenAI, DeepSeek and Moonshot did not immediately respond to requests for comment. Moonshot’s launch post attributed Kimi K3.1 Pro’s gains to increased post-training compute and large-scale reinforcement learning; DeepSeek’s update note, as described in the source material, did not detail its training data. Those statements do not independently resolve the allegation.
South Korea’s SBS reported additional details attributed to sources and quoted an OpenAI spokesperson saying the company takes the issue seriously and continues working on it. The original report’s paywall means those details cannot be checked against its full account here. They should be treated as reported claims, 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 reported use of Astra outputs is substantiated, the dispute would concern how a provider’s model outputs were used to develop competing systems, not whether distillation is inherently improper. Distillation is a standard training technique: a smaller model learns from answers produced by a larger one. The issue would be whether the outputs came from a rival model at significant scale and whether the way they were obtained complied with its terms.
The stakes for companies using AI include model provenance, contract risk and reliability. A customer may want to know what data or outputs shaped a model, particularly if a dispute could affect access, service terms or future availability. The report does not establish that users of the named models face a specific legal or operational consequence. The source material says their apps remain unchanged for now, and advises companies to check provenance, keep alternatives available and test models on their own tasks.
Price and benchmark comparisons also need care. The source material 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. On an unnamed independent index, it gives scores of 57, 46 and 32, respectively. These figures suggest lower listed prices alongside lower scores on that index; they do not show that the models are equivalent, or establish why they perform differently.
Products Worth Considering
Bose QuietComfort Wireless Noise-Cancelling Over-Ear Headphones, Black
LG 34GX90SA-W 34-inch Ultragear WQHD OLED Curved Gaming Monitor
Samsung Galaxy Tab S10 FE 128GB Wi-Fi Android Tablet, Blue
As an affiliate, we earn on qualifying purchases.
How Distillation Became the Dispute
In distillation, a teacher model generates answers that are used to train a smaller student model to imitate them. The method is used within AI labs to create more affordable or efficient models. A U.S. government advisory cited in the source material describes distillation as a legitimate and useful research technique. Its use alone does not show that a rule was broken.
The source material says OpenAI’s terms prohibit using its outputs to build competing models. That makes the allegation primarily a question of contractual compliance, while any alleged use of fake or resold accounts could raise separate questions about how access was obtained. Whether either occurred is not established by the public reporting described here. A terms-of-service allegation is not, by itself, a finding of illegal conduct.
There is earlier public context, but it does not prove the current claim. In September, OpenAI said a July campaign to extract hidden reasoning from its models was linked to a “core cluster” of individuals associated with Moonshot AI. The source material also says U.S. security agencies named DeepSeek and Moonshot in an advisory, while China’s commerce ministry called the accusations baseless. Those past statements and disputes are distinct from the new report about Astra outputs.
“Astra is “the most powerful model the company has released,” and one growing customer base is Chinese AI labs seeking to extract its capabilities.”
— The Information, in the public opening of its October 28 report
Evidence and Training Details Remain Sparse
The core allegation remains unconfirmed publicly. The material available here relies on The Information’s reporting, which cites people with direct knowledge, and Reuters’ summary. The companies’ responses were not available in Reuters’ account. Without the full report or supporting records, readers cannot assess the underlying evidence in detail.
It is also unclear how many Astra outputs were involved, whether hidden reasoning transcripts were accessed, what accounts or access routes were used, and how much any such data influenced the finished models. The launch materials cited for Kimi and DeepSeek do not answer those questions. The earlier OpenAI and government statements are relevant background, but they do not independently establish the new allegation.
No legal finding is reported in the source material. Whether conduct violated a contract, involved improper access or breached a law would depend on facts that have not been publicly set out. The comparison of model prices and scores likewise cannot establish how the models were trained or whether one copied another.
Company Responses and Evidence to Watch
The next useful development would be direct responses from OpenAI, DeepSeek and Moonshot, alongside more detail about the evidence, scale and means of access described in the report. Any further reporting should be distinguished from statements the companies make themselves, and from independent documentation or findings.
For businesses evaluating these models, the immediate step is to check current service terms, seek information about model provenance where it matters, and test performance on their own workloads. The source material also recommends keeping a fallback option and avoiding unofficial API resellers. Those are precautionary measures, not evidence that the models are unavailable or that users must stop using them.
For users, the report does not describe a change to the named apps or provide a confirmed action users need to take. The story remains a developing report: its significance will depend on whether the companies respond and whether more verifiable information clarifies what happened.
Key Questions
Which models are named in the report?
The report names Moonshot’s Kimi K3.1 Pro, released October 23, and DeepSeek-V4.1-Flash-1017, released October 17.
Have the companies confirmed that Astra outputs were used?
No. Reuters reported that OpenAI, DeepSeek and Moonshot did not immediately respond to requests for comment. The public material described here does not contain a confirmation or denial from any of them.
Is model distillation itself improper?
Not inherently. Distillation is a common training technique. The dispute is about the alleged use of a competitor’s outputs at scale and whether access and use complied with the provider’s terms.
Does the report establish that a law was broken?
No. The material reports an allegation, not a legal finding. It does not provide enough public detail to determine whether any contract or law was violated.
Do users need to change how they use the models now?
The source material reports no immediate change to the apps and identifies no confirmed action required of users. Companies can review terms and provenance, test models themselves and avoid unofficial API resellers while the reporting develops.
Source: ThorstenMeyerAI.com
