🔍 Read the full analysis: China’s New AI Models And GPT-6 Astra: What’s In The Distillation Report? 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 that DeepSeek and Moonshot AI relied heavily on OpenAI’s GPT-6 Astra outputs when training recent models. Reuters said the three companies did not immediately respond; public reporting has not established how the outputs were obtained or how much they affected either model.
The Information reported October 28 that Chinese AI labs DeepSeek and Moonshot AI relied heavily on outputs from OpenAI’s GPT-6 Astra to train recent models, including Moonshot’s Kimi K3.1 Pro and DeepSeek’s V4.1 Flash update. The allegation, attributed to people with direct knowledge, has not been publicly confirmed by any of the three companies, and key details about the training and access methods remain undisclosed.
Reuters summarized The Information’s account, naming Kimi K3.1 Pro, released October 23, and DeepSeek-V4.1-Flash-1017, launched October 17. The Information said both models benefited from distilling Astra, which was released in early September. Its full report is paywalled, and the public opening describes Chinese AI labs as a growing customer base seeking to extract Astra’s capabilities.
Reuters reported that OpenAI, DeepSeek and Moonshot did not immediately respond to requests for comment. That means the central claim remains an attributed report, not an independently established account in the material available publicly. The reporting does not provide figures for how many Astra outputs were used, which prompts were sent, or how the labs obtained access.
The labs have offered other explanations for their models’ development. Moonshot’s Kimi launch post credits its gains to substantially larger post-training compute and large-scale reinforcement learning. DeepSeek’s update note does not describe its training data. Neither note, as presented in the source material, addresses the allegation about Astra outputs.
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 dispute could affect how AI companies protect and license their systems, and how customers evaluate models built with them. OpenAI’s terms prohibit using its outputs to build competing models, according to the source material. Whether any specific conduct violated those terms would depend on facts that have not been made public, including how access was secured and what the labs did with the outputs.
The source compares prices and performance to show the commercial stakes, while cautioning that the models are not equivalent. It 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 the same independent index cited by the source, the scores are 57, 46 and 32 respectively. Those figures do not establish that distillation caused the price or score differences.
For companies choosing AI systems, the report is a reason to ask vendors about training-data provenance and model governance, and to test products on their own tasks. The source also advises maintaining a fallback and avoiding cheap, unofficial API resellers. For users, it says there is no reported change to the apps today; the larger issue is whether providers can protect model outputs and enforce their usage rules.
Products Worth Considering
Apple MacBook Pro 16-inch (2021) – M1 Pro Chip, 10-Core CPU, 16-Core GPU, 16GB RAM, 1TB SSD, Space Gray
Apple MacBook Pro 2023 14 with M3 Pro Chip, 18 GB RAM, 2 TB SSD, Space Black
Apple MacBook Pro 14 (2021) with M1 Max, 64 GB RAM, 2 TB SSD, Silver
As an affiliate, we earn on qualifying purchases.
How Distillation Became a Dispute
Distillation is a standard training technique: a smaller model learns to imitate responses generated by a larger “teacher” model. Labs commonly use their own systems this way to produce more compact or less expensive models. The U.S. government advisory cited in the source describes distillation as “a legitimate and useful technique in AI research.” The controversy is not the technique alone, but the claim that one company’s model outputs were used to train a rival’s system at scale, potentially in breach of service terms.
The source also describes an earlier dispute. In September, OpenAI linked a “core cluster” of a July campaign to extract hidden reasoning from its models to people associated with Moonshot AI. It says U.S. security agencies named both 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 latest training allegation.
A Korean broadcaster, SBS, reported additional details on October 29, saying the labs had previously focused on Anthropic’s Claude and shifted toward OpenAI after Anthropic tightened defenses. SBS also quoted an OpenAI spokesperson as saying the company takes the issue seriously and continues working on it. The source writer said those details could not be checked against The Information’s paywalled story.
What the Public Record Leaves Open
The central allegation has not been confirmed publicly by OpenAI, DeepSeek or Moonshot in the reporting cited here. The available account does not show the volume of Astra outputs involved, the prompts or methods used, whether hidden reasoning was obtained, or how much any such material contributed to the released models. It also does not establish whether access came through ordinary accounts, resellers, or another route.
Those gaps matter to both the technical and legal questions. Similar performance or lower prices would not, on their own, prove that a model was distilled from Astra. And a reported breach of terms is not a legal finding. The source’s account of possible contract issues and access fraud is analysis, not a conclusion from a court or regulator. The models’ launch materials cite different training explanations, but they do not resolve the allegation.
Responses and Evidence to Watch
The next clear developments would be responses from OpenAI, DeepSeek and Moonshot, followed by any additional reporting that provides evidence about training data, account access or the scale of the alleged activity. A company statement, technical disclosure, or regulatory action could clarify parts of the story; none is confirmed in the source material as pending or scheduled.
Until then, the report should be read as a serious but unresolved allegation. Businesses assessing the models can ask vendors for information about data provenance, terms and safeguards, and conduct their own evaluations rather than treating the report as proof of a particular model’s capabilities or origins. There is no reported change to the consumer apps in the source material, and no verified finding that users’ accounts or data were affected.
Key Questions
What does the report allege?
The Information reported that DeepSeek and Moonshot AI relied heavily on outputs from OpenAI’s GPT-6 Astra to train recent models. The claim is attributed to people with direct knowledge and has not been publicly confirmed by the three companies in the cited reporting.
Which models are named?
The report names Moonshot’s Kimi K3.1 Pro, released October 23, and DeepSeek-V4.1-Flash-1017, launched October 17.
Is distillation itself improper?
No. Distillation is a common model-training technique and can be used to create smaller versions of a lab’s own systems. The reported dispute concerns alleged use of a rival’s outputs and whether access or use breached the provider’s terms.
Have the companies confirmed the allegation?
No public confirmation is included in the source material. Reuters reported that OpenAI, DeepSeek and Moonshot did not immediately respond to its requests for comment.
Does the report change anything for users now?
The source reports no immediate change to the apps and no verified effect on user accounts or data. The allegation concerns model training; details about how the outputs were obtained and used remain unclear.
Source: ThorstenMeyerAI.com
