🔍 Read the full analysis: GPT-6 Astra, Distillation, And China’s Latest AI Models: A Closer Look on ThorstenMeyerAI.com
Get everyday 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 GPT-6 Astra outputs to train recent models, including Kimi K3.1 Pro and DeepSeek-V4.1-Flash-1017. Reuters said the three companies did not immediately respond to requests for comment; public evidence on the alleged methods and scale remains limited.
The Information reported Oct. 28 that Chinese AI labs DeepSeek and Moonshot AI relied heavily on outputs from OpenAI’s GPT-6 Astra to train recently released models. The report names Moonshot’s Kimi K3.1 Pro and DeepSeek’s V4.1-Flash-1017, but the companies have not publicly confirmed the account, and key details remain unavailable.
Reuters summarized The Information’s report, which cited people with direct knowledge. It said Kimi K3.1 Pro, released Oct. 23, and DeepSeek-V4.1-Flash-1017, launched Oct. 17, “benefitted from distilling Astra.” The Information’s public article opening described Chinese AI labs as a growing customer base for Astra; the rest of its report is behind a paywall, limiting what can be independently assessed from public material.
Reuters reported that OpenAI, DeepSeek and Moonshot did not immediately respond to its requests for comment. A separate Oct. 29 summary by South Korea’s SBS attributed a statement to an OpenAI spokesperson saying the company takes the issue seriously and continues working on it. SBS also reported that sources described a shift by Chinese labs from Claude to Astra after Anthropic tightened defenses. Those details could not be checked against The Information’s full report and should be treated as separately attributed reporting, not independently established facts.
The available accounts do not specify how many Astra outputs were used, how those outputs were obtained, or what share of either model’s training they represented. Moonshot’s Kimi launch post credited its gains to substantially larger post-training compute and large-scale reinforcement learning. DeepSeek’s update note did not describe its training data. Neither statement resolves 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 reported use is substantiated, the dispute could affect how AI companies protect and license model outputs, and how customers judge the origins of systems they buy. The distinction is not simply whether one model learned from another: distillation is a standard technique. The central questions are whether a rival’s outputs were used at scale, whether access complied with the provider’s terms, and whether accounts or other access routes were misrepresented.
Pricing and benchmark results give the allegation commercial significance but do not prove it. The source material lists Astra 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. It also gives scores of 57, 46 and 32 on the same independent index. The figures suggest lower listed prices and lower scores for the two Chinese models on that index; they do not establish how either model was trained or how it performs on a particular customer’s work.
For businesses, the report is a prompt to ask vendors about training-data provenance, terms and model evaluation, while maintaining an alternative provider where reliability matters. For users, the source says there is no immediate change to the apps. It advises avoiding cheap, unofficial API resellers, although the report does not establish that such resellers were involved in the alleged training.
Products Worth Considering
Samsung 27 Odyssey OLED G5 (G50SF) QHD QD-OLED Gaming Monitor
Logitech MX Keys S Wireless Keyboard, Graphite
Apple Watch SE 3 GPS 44mm Smartwatch
As an affiliate, we earn on qualifying purchases.
Distillation and Earlier Allegations
In model distillation, a smaller “student” model is trained on examples generated by a larger “teacher” model. AI developers commonly use the technique to make systems 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 dispute in this case concerns the alleged use of a competitor’s outputs and the way those outputs may have been obtained, not distillation by itself.
The allegation comes amid earlier public disputes over access to AI models. The source material says that in September OpenAI linked a “core cluster” in a July campaign to extract hidden reasoning from its models to individuals 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. Those claims and responses form background, but they do not independently verify the new allegation about Astra or establish that either named model was trained using it.
The source also cites an Anthropic experiment in which a model trained on full reasoning transcripts from Claude Opus 4.8 scored 80.4% on SWE-bench Verified, compared with 71.8% for a model trained on summarized transcripts. That is Anthropic’s own test, not evidence about Astra or the two models named in the report. It illustrates why access to detailed model outputs may matter to developers seeking to reproduce capabilities.
“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 Allegation
The claim remains unverified publicly. The Information cited people with direct knowledge, but the public material supplied here does not include the underlying evidence or the full paywalled reporting. Reuters said the companies did not immediately comment. There is no public account in the source material of the number of prompts or outputs involved, the accounts used, or the proportion of either model’s training that may have come from Astra.
It is also unclear whether the alleged activity involved hidden reasoning, ordinary answers, or both, and whether any access breached OpenAI’s terms. The source says OpenAI’s terms forbid using its outputs to build competing models, but determining a breach would depend on facts not available publicly and potentially on legal interpretation. The report, by itself, does not establish unlawful conduct.
Company Responses and Verification
The next developments to watch are direct responses from OpenAI, DeepSeek and Moonshot, any fuller disclosure of the reporting’s evidence, and whether regulators or other agencies issue statements tied specifically to this allegation. No enforcement action or formal legal finding is described in the source material.
Until more information is public, the report should be treated as an attributed allegation rather than a settled account of how Kimi K3.1 Pro or DeepSeek-V4.1-Flash-1017 were trained. Companies evaluating the models can test them on their own tasks and ask vendors about data provenance and access policies; readers should distinguish those practical precautions from proof that the reported conduct occurred.
Key Questions
What does the report allege?
The Information reported that DeepSeek and Moonshot relied heavily on GPT-6 Astra outputs to train recent models, naming Kimi K3.1 Pro and DeepSeek-V4.1-Flash-1017. The companies have not publicly confirmed the allegation in the cited reporting.
What is distillation?
Distillation trains a student model using examples or outputs from a larger teacher model. It is a common AI development technique; this report’s concern is the alleged use of a competitor’s outputs and whether access complied with the provider’s rules.
Has OpenAI confirmed that its terms were violated?
No violation is established in the source material. OpenAI’s terms are described as barring the use of its outputs to build competing models, but the public reporting does not provide enough detail to determine how outputs were obtained or whether the terms were breached.
Did the report change access to the models?
No immediate change to the apps is reported. The source offers no notice of a service suspension, product change or enforcement action related to the allegation.
What remains unknown?
The public account does not establish the amount of Astra output used, the methods of access, or the role it played in either model’s training. Responses from the companies and additional supporting evidence could clarify those points.
Source: ThorstenMeyerAI.com
