🔍 Read the full analysis: Distillation And China’s AI Models: Exploring The GPT-6 Astra Question on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- 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 OpenAI’s GPT-6 Astra outputs to train recent models. The claim is based on unnamed sources and has not been confirmed by the companies; the public record does not detail the alleged methods or scale.
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. The allegation concerns DeepSeek’s V4.1 Flash update and Moonshot’s Kimi K3.1 Pro, but the companies have not publicly confirmed it, and the available reporting gives no figures or detailed account of how the outputs were obtained.
Reuters summarized The Information’s report, which cited people with direct knowledge. It said Kimi K3.1 Pro and DeepSeek-V4.1-Flash-1017 “benefitted from distilling Astra.” Moonshot released Kimi K3.1 Pro on October 23, while DeepSeek launched its V4.1 Flash update on October 17, according to the supplied source material.
OpenAI, DeepSeek and Moonshot did not immediately respond to Reuters, the news service reported. The Information’s public article opening described demand for Astra and said Chinese AI labs were among customers seeking to extract its capabilities; the rest of the article is paywalled. The underlying evidence, quantities of model outputs, and alleged collection methods are not available in the supplied material.
The companies’ own release notes provide different explanations or no explanation: Moonshot attributed K3.1 Pro’s gains to “substantially larger post-training compute” and large-scale reinforcement learning, while DeepSeek’s update note did not specify training data, according to the source article. Those statements do not independently settle whether Astra outputs were 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
Why Training-Data Provenance Matters
If the report is accurate, it would sharpen a dispute over how far a lab can go in learning from a competitor’s model. Distillation is a standard training method, but using another provider’s outputs at scale may breach that provider’s contractual restrictions. The source material says OpenAI’s terms prohibit using its outputs to build competing models; whether any conduct violated those terms would depend on facts and legal interpretation that are not yet public.
The issue also has a practical cost dimension. The source article lists Astra at $10 per million input tokens and $50 per million output tokens, compared with Kimi K3.1 Pro at $2 and $10, and DeepSeek’s model at $0.30 and $1.20 at peak hours. It says the three score 57, 46 and 32 respectively on the same independent index. These figures suggest lower listed prices do not mean equivalent measured performance; they do not establish that distillation caused any particular price or score.
For businesses, the report is a reminder to ask vendors about training-data provenance and to test models on their own tasks rather than rely solely on general rankings. For users, the source article says there is no immediate change to the apps, while advising caution with inexpensive, unofficial API resellers. The report itself does not establish that ordinary users’ accounts or data are affected.
Products Worth Considering
Bose QuietComfort Wireless Bluetooth Headphones with Active Noise Cancellation, Mic, USB‑C Charging, Deep Bass, 24‑Hour Battery – White Smok
Bose QuietComfort True Wireless Earbuds with Active Noise Cancellation, Black
Apple iPad Air (4th Generation, 10.9-inch, Wi-Fi, 256GB) – Space Gray
As an affiliate, we earn on qualifying purchases.
Distillation and Earlier Allegations
In distillation, a large model—the “teacher”—produces answers that are used to train a smaller “student” model to imitate them. AI labs commonly use the technique with their own systems to make less expensive models. The source material quotes a U.S. government advisory describing distillation as “a legitimate and useful technique in AI research.” The controversy here is not the method by itself, but the reported use of a rival’s outputs and the terms and means under which those outputs may have been collected.
The report arrives amid earlier disputes over access to model outputs. The source article says OpenAI linked a “core cluster” of a July campaign to extract hidden reasoning from its models to people associated with Moonshot AI; it also says U.S. security agencies named both DeepSeek and Moonshot in an advisory. China’s commerce ministry called the accusations baseless, according to that material. These are separate claims and responses, not proof of the current allegation.
The source article also recounts earlier allegations involving attempts to obtain model responses at scale, but the supplied excerpt ends before giving the full account. Its author says a Korean broadcaster, SBS, reported additional details about a shift from targeting Anthropic’s Claude to OpenAI’s models and quoted an OpenAI spokesperson saying the company takes the issue seriously. The author says those details could not be checked against The Information’s paywalled 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 allegation remains unverified in the public record described here. The Information cited unnamed sources, and its full report is paywalled. The supplied material contains no estimate of how many Astra outputs were involved, how they were obtained, or how much they contributed to either model’s training.
It is also unclear whether any outputs came through ordinary paid access, resold or fake accounts, or another route. No public technical evidence is supplied that would establish the models’ training data or demonstrate that particular capabilities came from Astra. The companies’ lack of an immediate response to Reuters is not confirmation of the claim.
Any contract or legal conclusion would require more information, including the terms in force, the accounts used, and the nature and scale of the training. The report alone does not establish fraud, a breach, or a government finding against either lab.
Responses and Evidence to Watch
The next developments to watch are public responses from OpenAI, DeepSeek and Moonshot, along with any additional reporting that provides verifiable detail about the alleged data collection and training. The source material records no immediate Reuters response from the three companies and does not identify a formal investigation or legal action arising from this report.
For now, the distinction is between a reported allegation and established evidence. Further documentation, technical analysis, or statements from the companies could clarify whether Astra outputs were used, how access was obtained, and whether the reported activity crossed contractual or legal boundaries. Until then, the extent and consequences of the alleged distillation remain unresolved.
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, citing people with direct knowledge. Reuters summarized the report, but the claim has not been publicly confirmed by the companies.
Which models are named?
The report names Moonshot’s Kimi K3.1 Pro and DeepSeek-V4.1-Flash-1017. The source material gives release dates of October 23 and October 17, 2026, respectively.
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 models. The dispute concerns the reported use of a rival’s outputs and whether the access or training violated applicable terms or laws.
Has either company confirmed the allegation?
No confirmation is provided in the source material. Reuters said OpenAI, DeepSeek and Moonshot did not immediately respond to its request for comment. That does not establish whether the allegation is true or false.
Does the report mean users need to change anything now?
The source article says there is no immediate change to the apps. It advises businesses to check model provenance and test services directly, and users to be cautious with unofficial API resellers. The report does not describe a confirmed impact on user accounts or data.
Source: ThorstenMeyerAI.com
