🔍 Read the full analysis: What The Report Shows About Distillation In China’s New AI Models 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 that DeepSeek and Moonshot AI relied heavily on OpenAI’s GPT-6 Astra outputs to train recently released models. Reuters repeated the report, but the three companies did not immediately respond, and public details about the alleged data, methods and scale remain limited.
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 recently released models, a claim Reuters summarized within hours. None of the three companies had confirmed or denied the report in responses cited by Reuters, and the public material available so far does not establish how the outputs were obtained or how extensively they were used.
The models named in the report are Moonshot’s Kimi K3.1 Pro, released October 23, and DeepSeek-V4.1-Flash-1017, released October 17. Reuters, citing The Information, said both models “benefitted from distilling Astra.” The Information’s article, by Jing Yang, Juro Osawa and Qianer Liu, was published October 28 at 6 p.m. Pacific time. Its publicly visible opening described Chinese AI labs as a growing customer base for Astra; the detailed report is paywalled.
The account rests on unnamed sources described as having direct knowledge, according to the report as summarized by Reuters. Reuters said OpenAI, DeepSeek and Moonshot did not immediately respond to requests for comment. A Korean broadcaster, SBS, also summarized the report on October 29 and attributed a statement to an OpenAI spokesperson saying the company takes the issue seriously and continues working on it. That additional detail has not been independently checked against the paywalled article in the source material.
The public record does not give the number or type of Astra outputs allegedly used, the means by which they were accessed, or the portion of either model’s training that they may represent. Moonshot’s Kimi launch post attributed its gains to substantially larger post-training compute and large-scale reinforcement learning. DeepSeek’s update note, as described in the source material, did not discuss training data. Those company statements do not independently confirm or refute the report.
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 Sources Matter
If the report is accurate, it would put a routine AI training method at the center of a dispute about whose model outputs can be used and under what conditions. Distillation trains a model to imitate answers from another model. It can lower the cost of building a capable system, but using a competitor’s outputs at scale may violate that provider’s terms, depending on the facts and applicable agreements.
The commercial stakes are visible in the prices cited in the source material: 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 cited by the source, the models scored 57, 46 and 32, respectively. These figures suggest lower listed prices alongside lower scores on that index; they do not show that one model is equivalent to another, nor do they establish why the models differ.
For companies choosing AI systems, the report is a reason to ask vendors about training-data provenance and account access, while testing models on their own workloads and keeping alternatives available. For users, the source material says nothing changes in the apps immediately. It advises caution with cheap, unofficial API resellers, but offers no finding that a particular reseller was involved in this case.
Products Worth Considering
Bose QuietComfort Wireless Bluetooth Headphones with Active Noise Cancellation, Mic, USB‑C Charging, Deep Bass, 24‑Hour Battery – White Smok
LG UltraGear 32GX850A-B 32-Inch 4K UHD Glossy OLED Gaming Monitor
Samsung 27-Inch Odyssey QD-OLED G8 4K 240Hz Gaming Monitor
As an affiliate, we earn on qualifying purchases.
Distillation and Earlier Allegations
Distillation is an established training technique: a larger “teacher” model generates responses, which a smaller “student” model is trained to imitate. AI labs use the method to make models that can be cheaper or faster to run. The U.S. government advisory referenced in the source material describes distillation as a legitimate and useful research technique. The controversy in this report is not about the technique by itself, but the allegation that labs used a rival’s outputs and the unresolved questions about scale and access.
The report follows earlier allegations and warnings. The source material says that in September OpenAI connected 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 DeepSeek and Moonshot in an advisory, while China’s commerce ministry called the accusations baseless. These are attributed positions and allegations, not independent proof of the new report’s claims.
SBS reported that the labs had previously focused on Anthropic’s Claude and shifted to OpenAI after Anthropic tightened defenses. Because the original report’s full text is paywalled and this detail is available here only through SBS’s summary, it should be treated as secondhand reporting, not a fully verified account.
“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 and Access Remain Unclear
The report has not been publicly substantiated with training records, output samples, account details or quantified estimates. It is unclear how many Astra responses were involved, whether hidden reasoning or ordinary answers were used, and whether the outputs came through direct subscriptions, resold accounts or another route. The report’s reliance on unnamed sources and the paywall also limits what readers can assess independently.
It is also unresolved whether any use breached OpenAI’s terms or other legal obligations. The source material says OpenAI’s terms prohibit using its outputs to build competing models, but whether that provision applies to the alleged conduct would depend on evidence and the relevant agreements. No regulator or court finding is cited. The companies’ silence in Reuters’ initial reporting is not confirmation of the allegations.
Company Responses and Evidence
The immediate next development would be public responses from OpenAI, DeepSeek and Moonshot, along with any further reporting that supplies details about the sources, access methods or scale alleged. The source material does not identify a scheduled investigation, enforcement action or formal company announcement.
For now, readers and organizations should distinguish the reported claim from verified facts: the models and publication dates are identified, but the alleged training relationship remains unconfirmed. Any later account should clarify whether the companies dispute the report, provide evidence about model development, or address the terms governing access to Astra.
Key Questions
What does the report allege?
The Information reported that DeepSeek and Moonshot AI relied heavily on GPT-6 Astra outputs when training DeepSeek-V4.1-Flash-1017 and Kimi K3.1 Pro. The companies have not publicly confirmed the claim in the responses cited by Reuters.
What is AI distillation?
Distillation trains a smaller model to imitate responses generated by a larger model. It is a standard technique; the dispute here concerns the alleged use of a rival company’s outputs and whether access or use violated its terms.
Has OpenAI confirmed the alleged use?
No. Reuters said OpenAI, DeepSeek and Moonshot did not immediately respond to its requests for comment. SBS later attributed a general statement about taking the issue seriously to an OpenAI spokesperson, but that does not confirm the specific allegations.
What evidence has been made public?
The publicly described reporting relies on unnamed sources. The source material does not provide output counts, training records, access details or a measure of how much the alleged data affected either model.
Does the report mean the models are equivalent to Astra?
No. The source material lists different prices and scores for the three models on one independent index. Those figures do not establish equivalence, explain the score differences or prove that distillation caused any particular result.
Source: ThorstenMeyerAI.com
