🔍 Read the full analysis: China’s Newest AI Models: What We Know About Possible GPT-6 Astra Distillation on ThorstenMeyerAI.com
Get business pricing on office and shipping supplies
- 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 outputs from OpenAI’s GPT-6 Astra when training recent models. Reuters reported that none of the three companies had responded to its requests for comment. The public evidence does not specify how the outputs were obtained or how much they shaped either model.
The Information reported on October 28 that Chinese AI companies DeepSeek and Moonshot AI relied heavily on OpenAI’s GPT-6 Astra outputs to train recent models, a claim that could raise questions about model providers’ terms and the sourcing of training data. Reuters relayed the report within hours; none of the three companies had confirmed or denied it in comments to Reuters.
The models identified in the reporting are Moonshot’s Kimi K3.1 Pro, released October 23, and DeepSeek-V4.1-Flash-1017, released October 17. Reuters summarized The Information’s claim as the labs having “relied heavily” on Astra outputs. The Information’s public article opening described Chinese AI labs as a growing customer base for Astra; the full article is paywalled, limiting what can be independently checked from the public account.
The report does not publicly provide the number of prompts or outputs involved, the collection method, or the degree of influence on either model’s performance. The article’s source material says the claims rest on people with direct knowledge, but does not identify them publicly. Reuters said OpenAI, DeepSeek and Moonshot did not immediately respond to its requests for comment. Those non-responses do not establish whether the allegation is accurate.
There is also no public confirmation that either company obtained outputs through fake accounts, resellers or another method. Moonshot’s launch post attributed Kimi K3.1 Pro’s gains to more post-training compute and large-scale reinforcement learning. DeepSeek’s October update note did not describe its training data. These statements offer some information about the companies’ accounts of their work, but do not independently resolve the reported use of 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 Sources Matter
If the report is substantiated, it could sharpen disputes over whether developers may use a rival’s model outputs to train competing systems. OpenAI’s terms, as described in the source material, prohibit using its outputs to build competing models. Whether a particular use breached those terms would depend on facts not yet public, including how access was obtained, what the terms said at the time, and what the labs did with the outputs. This is a report about a possible contractual dispute, not a finding of illegality.
The issue also matters to businesses selecting AI services. The source article 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 gives scores of 57, 46 and 32 respectively on an unspecified independent index. Those figures suggest different price and score profiles in the cited comparison; they do not show that the models are interchangeable, nor do they establish how either Chinese model was trained.
For users, the report does not describe an immediate change to the apps or services. For organizations, it is a reason to ask vendors about training-data provenance, access controls and terms compliance, and to test models on their own workloads rather than infer quality or reliability from a single benchmark.
Popular right now
De'Longhi Magnifica Evo Automatic Espresso Machine with Milk Frother – Silver
Breville BES870XL Barista Express Espresso Machine, Brushed Stainless Steel
De'Longhi Magnifica Evo Fully Automatic Espresso Machine with Manual Milk Frother
As an affiliate, we earn on qualifying purchases.
Earlier Disputes Over Model Outputs
Distillation is a standard machine-learning technique: a smaller model is trained using answers generated by a larger “teacher” model. The U.S. government advisory cited in the source calls it “a legitimate and useful technique in AI research.” Labs can use their own models this way. The contested issue in this report is the alleged use of a competitor’s outputs at scale, potentially in ways that violate service rules.
The allegation follows earlier public disputes about efforts to extract model capabilities. The supplied source says OpenAI linked a July campaign aimed at extracting hidden reasoning to people associated with Moonshot AI in a September statement, and that U.S. security agencies named both DeepSeek and Moonshot in an advisory. It also says China’s commerce ministry called accusations baseless. These are separate claims and responses; they do not prove the new allegation about Astra. The source further describes a February Anthropic account involving large volumes of exchanges with Claude, but the provided text cuts off before giving full details, so that episode cannot be fully assessed here.
A related account from South Korea’s SBS, as summarized in the supplied material, said sources told the broadcaster Chinese labs had previously focused on Claude and shifted toward OpenAI after Anthropic tightened defenses. SBS also attributed a statement to an OpenAI spokesperson saying the company takes the issue seriously. Those details have not been checked against The Information’s full article in the material available for this 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 Still Missing From Public View
The central claim remains unverified by public statements from the companies involved. The available account does not say how many Astra outputs were used, whether they included hidden reasoning, which prompts generated them, or how much they affected the final models. The Information’s full report is paywalled, and Reuters’ summary does not answer these questions.
It is also unclear whether any access involved fake, shared or resold accounts, or whether the reported activity violated terms in force at the time. The supplied material notes that such methods are part of the wider concern about industrial-scale distillation, but does not establish that DeepSeek or Moonshot used them in this case. No legal or regulatory finding is described.
What Could Clarify the Allegation
The next meaningful developments would be responses from OpenAI, DeepSeek or Moonshot, further reporting that supplies verifiable details, or public documentation from the companies about training data and account access. Any subsequent investigation or enforcement action would also need to establish the relevant facts before conclusions about contract violations could be drawn.
Until then, the report should be treated as an allegation attributed to unnamed sources, not as a confirmed account of how the models were built. Companies evaluating the systems can review vendor disclosures and service terms, run their own task-specific tests, and keep alternatives available. For consumers, the supplied reporting describes no immediate service change; the key questions remain the evidence behind the claim and the companies’ responses.
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: DeepSeek-V4.1-Flash-1017 and Moonshot’s Kimi K3.1 Pro. The companies have not publicly confirmed the claim in the cited Reuters account.
Does the report prove that DeepSeek or Moonshot broke the law?
No. The public reporting does not establish a legal violation. OpenAI’s terms reportedly bar using its outputs to build competing models, but whether those terms were breached would depend on details that remain unknown and unverified.
What is AI model distillation?
Distillation trains a smaller “student” model using answers produced by a larger “teacher” model. It is a recognized technique; the dispute here concerns the alleged use of a competitor’s outputs and whether that use complied with the provider’s terms.
Have the companies responded to the allegation?
Reuters reported that OpenAI, DeepSeek and Moonshot did not immediately respond to its requests for comment. The supplied material does not include a direct confirmation or denial from any of them about the specific Astra claim.
Will this report change how users access the models?
The supplied reporting describes no immediate change to the apps or services. It does not establish that either model is unavailable or that users need to change their settings; further company statements or reporting may clarify the issue.
Source: ThorstenMeyerAI.com
