🔍 Read the full analysis: Could Distillation Connect China’s New AI Models To GPT-6 Astra? on ThorstenMeyerAI.com
Get hardware and tech essentials 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 outputs from OpenAI’s GPT-6 Astra to train recent models. Reuters summarized the report, but the companies did not respond to its requests for comment, and public evidence about the scale or method remains limited. Distillation is a common training technique; the central questions are whether a rival model’s outputs were used at scale and whether access violated OpenAI’s terms.
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, including DeepSeek-V4.1-Flash-1017 and Moonshot’s Kimi K3.1 Pro. Reuters summarized the report, but none of the three companies had confirmed or denied the central claim in response to Reuters’ requests, leaving the scale and method of any alleged use unclear.
The Information’s report, by Jing Yang, Juro Osawa and Qianer Liu, cited people with direct knowledge, according to Reuters. The Information’s public opening described Astra as a powerful OpenAI model released in early September and said Chinese AI labs were among a growing customer base seeking to extract its capabilities. The rest of the article is behind a paywall, limiting what can be independently assessed from the public account.
Reuters identified the models named in the report as DeepSeek-V4.1-Flash-1017, released Oct. 17, and Kimi K3.1 Pro, released Oct. 23. Reuters said the report described both as having benefited from distillation of Astra. It also reported that OpenAI, DeepSeek and Moonshot did not immediately respond to its requests for comment. The companies’ silence does not establish whether the allegation is accurate.
A separate account by South Korea’s SBS on Oct. 29 attributed further details to the report, including an assertion that Chinese labs had shifted from using Claude outputs after Anthropic tightened defenses. SBS also quoted an OpenAI spokesperson as saying the company takes the issue seriously and continues working on it. Those details could not be checked against the paywalled reporting in the supplied source material. Moonshot’s launch post credited Kimi K3.1 Pro’s gains to increased post-training compute and large-scale reinforcement learning; DeepSeek’s update note did not describe its training data.
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 for Model Training
The allegation matters because distillation can let a smaller model learn from a more capable model’s answers, potentially reducing the time and computing resources needed to build a competing system. The technique itself is widely used: labs can train smaller versions of their own models using outputs from larger ones. The dispute is about whether a lab used a rival’s outputs at significant scale and whether those outputs were obtained in ways that breached access rules.
OpenAI’s terms prohibit using its outputs to build competing models, according to the source material. If the reported conduct occurred, the dispute could involve contractual restrictions and questions about how accounts or access were used. Whether any conduct was unlawful is a separate legal question; the available reporting does not establish that a law was broken.
The commercial stakes are visible in the prices and benchmark figures cited by the source. GPT-6 Astra is listed 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 the same independent index, the models scored 57, 46 and 32, respectively. Those figures suggest lower listed prices alongside lower scores on that index, not equivalent performance. The source does not identify the index or provide enough detail to generalize the results to every task.
For organizations choosing models, the report is a reminder to ask about training-data provenance, contractual terms and independent testing. For users, it does not establish that the named apps or models are unsafe or that anything has changed in their operation. The source advises caution with unofficial, low-cost API resellers, whose access and data practices may be difficult to verify.
Products Worth Considering
Apple MacBook Pro 16.2-inch with M5 Pro, 18-core CPU, 20-core GPU, 48GB Memory, 1TB SSD, Wi-Fi 7, Space Black
Apple MacBook Pro 16-inch (2021) – M1 Pro Chip, 10-Core CPU, 16-Core GPU, 16GB RAM, 1TB SSD, Space Gray
Apple MacBook Pro (14.2-inch, 2025) – M5 chip, 16GB Unified Memory, 1TB SSD, Space Black
As an affiliate, we earn on qualifying purchases.
How Distillation Became a Dispute
In distillation, a teacher model generates answers to prompts, and a student model is trained to imitate them. The method can transfer useful capabilities into a less expensive model. The U.S. government advisory cited in the source calls distillation a legitimate and useful technique in AI research. Its legitimacy as a method does not settle whether a particular use complies with a provider’s terms.
The source also describes earlier concerns about attempts to extract models’ hidden reasoning. In September, OpenAI linked what it called a “core cluster” of a July campaign to extract hidden reasoning to individuals associated with Moonshot AI, and U.S. security agencies named both DeepSeek and Moonshot in an advisory. China’s commerce ministry called the accusations baseless. These are attributed claims and responses, not independent confirmation of the new report’s allegation.
The supplied account refers to an earlier Anthropic statement that DeepSeek, Moonshot and MiniMax generated more than 16 million exchanges with Claude through accounts. That history provides context for increased scrutiny of model-output collection, but it does not by itself prove that DeepSeek or Moonshot used Astra outputs to train the models named in the latest report.
“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 public record remains limited. The Information’s report relies on unnamed sources, and its full account is paywalled. The public summaries do not specify how many Astra outputs were collected, which prompts were used, how access was obtained, or how much those outputs contributed to either model’s training. No company had publicly confirmed the central allegation in the reporting described here.
It is also unclear whether the models were trained directly on Astra answers, whether any outputs came through authorized accounts, or whether the reported activity violated OpenAI’s terms. A claim of contract infringement would require evidence about the terms, the conduct and the relevant accounts; the available material does not resolve those issues. The model makers’ own launch explanations provide some context but do not independently establish or rule out distillation from Astra.
Responses and Evidence to Watch
The next concrete developments would be public responses from OpenAI, DeepSeek or Moonshot, or further reporting that provides verifiable details about the alleged data collection and training process. More information about account access, the volume of outputs and how they were used would help readers distinguish ordinary model use from large-scale extraction claimed to breach terms.
Companies evaluating AI systems can review their own supplier and training-data disclosures, check whether API providers are authorized, and test models on tasks relevant to their work rather than relying on a single benchmark. The report does not establish an immediate change for users of the named products. Any legal or policy consequences will depend on evidence and on how the companies and relevant authorities respond.
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. Reuters summarized the report; the companies had not confirmed the claim in the reporting described here.
Which models were named?
The report named DeepSeek-V4.1-Flash-1017, released Oct. 17, and Kimi K3.1 Pro, released Oct. 23, according to Reuters’ summary.
Is distillation itself improper?
No. Distillation is a common training method, including when a lab uses its own model’s outputs to train a smaller one. The concern in this report is the alleged use of a rival’s model outputs at scale and whether the way they were obtained complied with the provider’s rules.
Has OpenAI confirmed the allegation?
No confirmation is reported. Reuters said OpenAI, DeepSeek and Moonshot did not immediately respond to its requests for comment. SBS separately quoted an OpenAI spokesperson discussing the issue generally, which does not confirm the specific allegation.
Does the report mean users need to change anything now?
The source material reports no immediate change to the named apps for users. It does not establish that the models are unsafe or that their performance claims are invalid. The allegation and its legal implications remain unverified.
Source: ThorstenMeyerAI.com
