🔍 Read the full analysis: GPT-6 Astra And China’s New AI Models: What The Distillation Report Does And Doesn’t Show on ThorstenMeyerAI.com
Get the little things that make your day 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 October 28 that DeepSeek and Moonshot AI relied heavily on outputs from OpenAI’s GPT-6 Astra to train recent models. Reuters reported that the three companies did not immediately respond, and the public evidence does not establish how the outputs were obtained or how extensively they were used. Distillation is a standard training technique; the dispute concerns whether a rival’s model was used in ways that violated OpenAI’s terms.
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, a claim that could raise questions about competition and compliance with OpenAI’s terms. Reuters said OpenAI, DeepSeek and Moonshot did not immediately respond; the available public reporting does not specify how the outputs were obtained or how much they shaped the models.
The Information’s report, by Jing Yang, Juro Osawa and Qianer Liu, said Astra had become a target for Chinese AI labs seeking to extract capabilities from the model. Reuters summarized the report as saying the labs’ latest models “relied heavily” on Astra outputs. The models identified were Moonshot’s Kimi K3.1 Pro, released October 23, and DeepSeek-V4.1-Flash-1017, released October 17.
The report’s detailed account is behind a paywall, and the publicly available material does not include figures for the volume of model outputs involved, the prompts used, or the route through which the labs accessed Astra. The source material says the three companies had not confirmed or denied the claim in responses to Reuters. Moonshot’s launch post attributed Kimi K3.1 Pro’s 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.
A Korean broadcaster, SBS, separately summarized details attributed to sources in The Information’s report, including that Chinese labs had shifted from targeting Anthropic’s Claude to OpenAI models after Anthropic strengthened defenses. The original account could not be independently checked against its paywalled reporting in the supplied material. SBS also quoted an OpenAI spokesperson as saying the company takes the issue seriously and continues working on it.
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
The Stakes for AI Model Access
If the reported use is substantiated, it would sharpen a dispute over who can learn from a commercial AI model’s outputs and under what conditions. Model distillation is widely used to make smaller or less expensive systems imitate a more capable model. Using a competitor’s outputs at scale, however, may conflict with the provider’s terms, particularly if access was obtained through deceptive accounts or unauthorized resellers. The available report does not establish that either lab used such methods.
The commercial incentives are clear from the prices and performance figures in the source material, though those figures do not prove the reported training relationship. 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, the three score 57, 46 and 32 respectively. These figures indicate lower listed prices for the two Chinese models, but also lower scores on that index—not equivalent performance at a lower price.
For businesses selecting models, the report is a reminder to ask vendors about training-data provenance and to test systems on their own workloads. It does not establish that the named models are unsafe or unusable. For ordinary users, the source says there is no immediate change to the apps; the practical caution offered is to avoid cheap, unofficial API resellers whose access and sourcing cannot be verified.
Products Worth Considering
Google Nest Learning Thermostat (4th Gen) with Nest Temperature Sensor (2nd Gen), Obsidian
eufy Security Essential Alarm System, 7-Piece, No Monthly Fee
iRobot Roomba 105 Vac Robot Vacuum
As an affiliate, we earn on qualifying purchases.
How Distillation Became a Dispute
Distillation is a normal training method: a smaller student model is trained on responses produced by a larger teacher model. AI labs often use their own systems this way to create more efficient versions. The U.S. government advisory cited in the source material describes distillation as “a legitimate and useful technique in AI research.” The ethical and contractual debate changes when the teacher is a competitor’s model and its outputs are gathered at significant scale.
OpenAI’s terms, according to the source material, prohibit using its outputs to build competing models. That makes the central issue potentially contractual, while any deception used to obtain access could raise additional questions. Whether any law was broken is not established by the reporting summarized here and would depend on facts and applicable law.
The report lands amid earlier allegations about model extraction. The source material says that in September OpenAI linked a “core cluster” of a July campaign to extract hidden reasoning from its models to people associated with Moonshot AI, and that U.S. security agencies named both labs in an advisory. China’s commerce ministry called the accusations baseless. Those earlier claims are a backdrop, not proof of the new report’s specific allegation about Astra or the two named models.
“DeepSeek and Moonshot “relied heavily on outputs” from OpenAI’s GPT-6 Astra to train their latest models.”
— The Information, as summarized by Reuters
What the Public Record Does Not Show
The report’s central allegation remains unconfirmed by the companies in the supplied material. It does not disclose how many Astra responses were used, which accounts generated them, whether access was direct or through intermediaries, or how much the outputs contributed to either model. Those details matter for judging both the scale of the activity and whether it violated service terms.
It is also unclear whether the companies will issue fuller responses, whether OpenAI has made a formal legal complaint, or whether any investigation is underway. The article’s account of earlier activity and the claims relayed by SBS should not be treated as independent confirmation of the latest allegation. No public evidence provided here establishes that the named models were trained primarily on Astra, or that their developers broke the law.
Responses and Evidence to Watch
The next developments to watch are responses from OpenAI, DeepSeek and Moonshot, along with any further reporting that provides evidence about the volume and source of the alleged outputs. Clarification from the labs about their training methods—or documentation from OpenAI about any account restrictions or enforcement action—could help establish what happened.
Until more information is available, companies evaluating AI vendors can ask about data provenance, terms of access and safeguards against unauthorized collection, while testing models against their own needs. The report itself does not change the models’ published prices or performance scores, and the supplied material describes no immediate change for users of their apps.
Key Questions
What did The Information report?
It reported that DeepSeek and Moonshot AI relied heavily on outputs from OpenAI’s GPT-6 Astra when training recent models, including DeepSeek-V4.1-Flash-1017 and Moonshot’s Kimi K3.1 Pro. The claim has not been confirmed by the companies in the supplied reporting.
What does distillation mean in AI?
Distillation trains a smaller model to imitate responses from a larger model. It is a common technique; the dispute in this report concerns the alleged use of a competitor’s outputs and whether they were gathered in a way that breached access terms.
Did DeepSeek or Moonshot break the law?
The available reporting does not establish that. OpenAI’s terms are described as barring use of its outputs to build competing models, but whether a contract was breached—or whether any law was broken—depends on facts not disclosed in the public material.
Have the companies responded?
Reuters said OpenAI, DeepSeek and Moonshot did not immediately respond to its requests. SBS separately quoted an OpenAI spokesperson as saying the company takes the issue seriously and continues working on it. The supplied material does not include a direct response from DeepSeek or Moonshot.
Does the report change anything for users right now?
The supplied material identifies no immediate change to the apps or services. It advises users to be cautious with cheap, unofficial API resellers, but the report does not establish that the named models are unsafe or that their availability has changed.
Source: ThorstenMeyerAI.com
