🔍 Read the full analysis: China’s AI Models And GPT-6 Astra: A Careful Read Of The Distillation Report on ThorstenMeyerAI.com
Get the latest gadgets 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 OpenAI’s GPT-6 Astra outputs to train recently released models. Reuters reported that the companies did not immediately respond; the public reporting gives no scale, method or independently verifiable evidence, so the allegation 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, including DeepSeek’s V4.1 Flash update and Moonshot’s Kimi K3.1 Pro. Reuters relayed the report, but the public account does not provide figures or a method, and none of the three companies had responded to Reuters at the time of its report.
The Information’s report, by Jing Yang, Juro Osawa and Qianer Liu, cited people with direct knowledge of the matter, according to Reuters. The report said Kimi K3.1 Pro and DeepSeek-V4.1-Flash-1017 benefited from distilling Astra. Moonshot released Kimi K3.1 Pro on Oct. 23; DeepSeek released its update on Oct. 17.
The original article’s 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 article is paywalled, and the public material does not specify how many Astra outputs were used, how those outputs were acquired, or how much they affected either model’s training.
Reuters said OpenAI, DeepSeek and Moonshot did not immediately respond to its request for comment. The supplied launch materials offer no confirmation of the reported training: Moonshot attributed Kimi’s gains to more post-training compute and large-scale reinforcement learning, while DeepSeek’s update note did not describe its training data. Those statements do not, by themselves, resolve the allegation.
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 report is accurate, the central issue is not the use of distillation by itself, but whether a lab used a rival’s model outputs at scale and in a way that breached the provider’s rules. OpenAI’s terms prohibit using its outputs to build competing models, according to the source material. Whether any particular conduct violated a contract or other law would depend on facts not disclosed in the public report.
The allegation also matters to companies choosing AI systems. Model performance, cost and training provenance can affect procurement, compliance review and the reliability of claims about how a model was built. The source material 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. It also reports independent-index scores of 57, 46 and 32, respectively. Those figures are presented without enough information here about the index’s methods or date to support broader conclusions; they do indicate that the cheaper models should not be treated as equivalent on that measure.
For users, the report does not establish that the models are unsafe or that anything in their apps has changed. For organizations, it is a reason to ask vendors about data and model provenance, test systems on their own tasks, and keep alternatives available rather than treating an allegation as proof.
Popular right now
De'Longhi Rivelia Fully Automatic Espresso Machine with Milk Frother, White
Vitamix Propel Series 750 Professional-Grade Blender, 4 Programs, Black
KitchenAid Artisan Series 5-Quart Tilt-Head Stand Mixer, Dried Rose (KSM150PS)
As an affiliate, we earn on qualifying purchases.
Distillation and Earlier Allegations
Distillation is a training method in which a student model learns from answers generated by a larger teacher model. Developers use it to make smaller or less costly systems imitate some of a larger model’s capabilities. The source material notes that a U.S. government advisory describes distillation as a legitimate and useful AI research technique. The method alone does not establish wrongdoing; the relevant questions include who owns the teacher model, how its outputs were obtained, and whether the use complied with applicable terms.
The report arrives after earlier disputes over access to U.S. AI systems. In September, OpenAI said it had linked a “core cluster” in a July campaign to extract hidden reasoning from its models to 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 prior statements provide context but do not independently prove the new claim about Astra or establish the conduct of either lab.
South Korean broadcaster SBS, in a recap dated Oct. 29, reported additional details attributed to sources and quoted an OpenAI spokesperson saying the company takes the issue seriously. The supplied account says those details could not be checked against The Information’s full report. They should be treated as secondary reporting, not as independently verified findings.
“one growing customer base”
— The Information, as quoted in the supplied report
What the Public Record Does Not Show
The allegation remains based on reporting attributed to unnamed sources, with the original article’s full account behind a paywall. The available material does not show the number or type of Astra outputs involved, how access was obtained, whether accounts were shared or resold, or what evidence supports the description “relied heavily.” It also does not establish whether any model’s final performance can be traced to Astra outputs.
No response from OpenAI, DeepSeek or Moonshot was available to Reuters when it published its account. It is not clear whether any company later issued a statement, whether OpenAI has taken action, or whether regulators have opened an inquiry. The earlier September allegations are separate and should not be treated as confirmation of the current report.
Company Responses and Evidence to Watch
The next useful developments would be direct statements from OpenAI, DeepSeek and Moonshot, along with any additional reporting that explains the evidence, scale and alleged access methods. A public account of a complaint, investigation or enforcement action would clarify whether the dispute has moved beyond reporting and contractual concerns. Until then, the claims should remain attributed rather than presented as established fact.
Organizations evaluating these models can review vendor disclosures, ask about training-data and model-output provenance, and test performance on their own workloads. Users should also distinguish official services from unofficial API resellers, as the source material advises. These steps address practical procurement and access risks without assuming the allegations have been proven.
Key Questions
What did The Information report?
It reported that DeepSeek and Moonshot AI relied heavily on GPT-6 Astra outputs to train DeepSeek-V4.1-Flash-1017 and Kimi K3.1 Pro. The claim is attributed to people with direct knowledge, and has not been publicly confirmed by the companies in the material provided.
What is distillation in AI?
Distillation trains a smaller student model using outputs from a larger teacher model. The method is used in AI development; the dispute here concerns the alleged use of a competitor’s outputs and whether the way they were obtained or used violated rules.
Have OpenAI, DeepSeek or Moonshot confirmed the report?
No. Reuters reported that all three did not immediately respond to its requests for comment. The supplied material does not include a later confirmation or denial.
Does the report prove that either model is unsafe?
No. The report concerns alleged training practices, not a finding that either model is unsafe. The public information provided does not establish the scale of any use or its effect on model behavior.
What should AI buyers do now?
Buyers can ask vendors about model and training-data provenance, evaluate systems on their own tasks, and keep alternatives available. The allegation alone is not proof of a violation or a reason to assume performance or safety claims are false.
Source: ThorstenMeyerAI.com
