🔍 Read the full analysis: China’s AI Models And GPT-6 Astra: What The Distillation Report Leaves Unanswered on ThorstenMeyerAI.com
Get smart everyday buys 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 that DeepSeek and Moonshot AI relied heavily on OpenAI’s GPT-6 Astra outputs to train recent models. Reuters said the companies involved did not respond to its requests for comment; the public reporting gives no figures or details about how the outputs were obtained. Distillation is a standard training method, but using a rival’s outputs may violate OpenAI’s terms, depending on the facts.
The Information reported on 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 the account, and the available reporting does not describe the scale or method of the alleged training.
Reuters summarized The Information’s report, which cited people said to have direct knowledge. It named Kimi K3.1 Pro, released October 23, and DeepSeek-V4.1-Flash-1017, released October 17, as models that benefited from distillation using Astra outputs. The Information’s public article opening described Chinese AI labs as a growing customer base for Astra; much of its reporting is behind a paywall.
Reuters said OpenAI, DeepSeek and Moonshot did not immediately respond to requests for comment. A separate account from South Korea’s SBS on October 29 attributed a statement to an OpenAI spokesperson 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, or a public confirmation from OpenAI that the specific models were trained this way.
The reporting available publicly does not provide the number of Astra outputs involved, the prompts used, or how access was obtained. Moonshot’s launch post credited Kimi K3.1 Pro’s gains to substantially greater post-training compute and large-scale reinforcement learning. DeepSeek’s update note, according to the source material, did not describe its training data. 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 allegation is substantiated, it could sharpen disputes over how AI companies may use one another’s systems to develop competing products. Distillation is widely used, including to make smaller or less expensive models imitate a more capable teacher model. The central issue is not the technique alone but whether a competitor’s outputs were collected at scale in ways barred by its service terms or obtained through deceptive access.
The report also matters to organizations choosing models. The source material 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 cites scores of 57, 46 and 32 respectively on the same independent index. These figures suggest lower prices do not mean identical measured performance; they do not establish why the models scored as they did or prove the alleged training relationship.
For business users, the immediate practical issue is model provenance and reliability, not a demonstrated change to the apps. Organizations can record where models come from, test them on their own workloads and keep alternatives available. The source advises users to avoid cheap, unofficial API resellers, though it does not identify any reseller implicated in this report.
Products Worth Considering
HP Sprocket Photo Booth Machine Instant Color Photo Printer with 10.1 Touchscreen, LED Ring Light, 250+ Frames, 3×4 ZINK Paper, QR Code Shar
Dell 24 Plus Monitor S2425HSM – 23.8 FHD (1920×1080) 144Hz 1ms, 2 x 3W Speakers, HDMI, Height/Tilt/Pivot/Swivel Adjustability, AMD FreeSync
Dell 34-Inch S3425DW Curved WQHD USB-C Monitor
As an affiliate, we earn on qualifying purchases.
Earlier Disputes Over Model Access
The report follows earlier allegations about attempts to extract model capabilities. The source material says OpenAI linked a July campaign to obtain hidden reasoning from its models to individuals associated with Moonshot AI, and that U.S. security agencies later named DeepSeek and Moonshot in an advisory. China’s commerce ministry called the accusations baseless. These are separate, attributed claims; they do not establish that the current models were trained on Astra outputs.
Distillation means training a student model on answers generated by a teacher model, so it can imitate some of the teacher’s capabilities. The U.S. advisory cited in the source material calls distillation “a legitimate and useful technique in AI research.” The point of contention arises when a lab uses another provider’s model at scale in a manner that may violate that provider’s terms. OpenAI’s terms, as described in the source, prohibit using its outputs to build competing models, making the question potentially a contractual one. Whether any conduct also involved fraud or other legal violations depends on facts not provided in the public account.
The source also describes 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% when trained on summarized transcripts. That is Anthropic’s own experiment, not evidence about the DeepSeek or Moonshot models. The source says SBS reported that Chinese labs had previously focused on Claude and shifted toward OpenAI after Anthropic tightened defenses, but that detail 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 Allegation
The public record described here does not show how many Astra outputs were used, which prompts produced them, whether hidden reasoning was accessed, or whether the alleged use was direct or through third parties. It also does not identify the accounts involved or provide underlying documents, technical analysis or a response from the two labs.
The claim rests on The Information’s reporting from unnamed sources, as summarized by Reuters. The full article is paywalled, and Reuters reported that the companies did not immediately respond. The distinction between ordinary distillation and a possible breach of terms depends on the specific access and training methods; the available information does not establish a legal violation or its consequences.
Company Responses and Further Evidence
The next developments to watch are direct responses from OpenAI, DeepSeek and Moonshot, along with any fuller account of the sourcing and methods behind the report. Technical documentation, account records or independent analysis could help establish whether Astra outputs were used and on what scale. No investigation, enforcement action or legal case tied specifically to this report is identified in the supplied material.
For now, the allegation should remain distinct from confirmed facts: the report names two models and says they benefited from Astra outputs, while the companies have not publicly confirmed that account in the source material. There is no reported change to the models’ availability or to user-facing services. Until more evidence or company statements emerge, questions about training provenance and compliance remain open.
Key Questions
Which models are named in the report?
The report names Moonshot’s Kimi K3.1 Pro, released October 23, and DeepSeek-V4.1-Flash-1017, released October 17. The claim that they benefited from Astra outputs has not been publicly confirmed by the companies in the supplied material.
What is AI distillation?
Distillation is a training method in which a student model learns from answers produced by a teacher model. It is a common technique; disputes may arise over whether the teacher model belongs to a competitor and whether its terms permit the use.
Has OpenAI confirmed the alleged use?
No public confirmation is included in the source material. Reuters reported that OpenAI, DeepSeek and Moonshot did not immediately respond to its requests for comment. SBS separately attributed a general comment to an OpenAI spokesperson.
Does the report prove that a law was broken?
No. The available reporting does not establish how the outputs were obtained or whether any conduct violated a contract or law. The question may involve OpenAI’s terms, but a legal conclusion would require more evidence.
Will users notice an immediate change?
The supplied material reports no change to the models’ apps or availability. The immediate implications are mainly for questions about model provenance, provider terms and how businesses evaluate AI services.
Source: ThorstenMeyerAI.com
