🔍 Read the full analysis: China’s New AI Models: A Report On Distillation, GPT-6 Astra, And Open Questions on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
The Information reported on October 28 that DeepSeek and Moonshot AI relied heavily on OpenAI’s GPT-6 Astra outputs to train recent models. Reuters said the three companies did not immediately respond, and public details about the alleged training method, scale and access remain unavailable. Distillation is a standard AI technique, but using a rival’s outputs could violate 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. Reuters repeated the report’s central claim, but said OpenAI, DeepSeek and Moonshot did not immediately respond; the public reporting does not establish how the outputs were obtained or how much they shaped either model.
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 summarized The Information’s report as saying the labs “relied heavily” on Astra outputs. The Information’s public opening described Chinese AI labs as a growing customer base for Astra, but its full article is behind a paywall. The detailed claim rests on sources cited by that publication, rather than public training records or a disclosed dataset.
No method or volume figures were provided in the material available publicly: it remains unknown how many Astra responses were used, which prompts were involved, or whether the labs accessed them directly, through resellers, or by another route. Moonshot’s release post attributed Kimi K3.1 Pro’s 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 report’s allegation.
Reuters reported that the three companies did not immediately respond to requests for comment. A separate summary by South Korea’s SBS on October 29 attributed a response to an OpenAI spokesperson saying the company takes the issue very seriously and continues working on it. That statement has not been independently checked against The Information’s full report in the source material provided.
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 Model Training
The allegation matters because distillation can transfer capabilities from a large model into a cheaper one. A lab prompts a “teacher” model and uses its answers to train a “student” model to imitate them. This is a common research and engineering method when labs train smaller versions of their own systems. The dispute is not about distillation itself, but whether a rival’s model was used at scale in a way that breaches its terms or involved improper access.
OpenAI’s terms, as described in the source material, prohibit using its outputs to build competing models. If the reported conduct is substantiated, the central issue could be contractual, with possible separate questions about how access was obtained. The public evidence does not establish a legal violation, and the source author’s interpretation is not a legal finding.
There may also be a commercial incentive. The source 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 the same independent index, without naming that index in the supplied material. Those figures suggest lower listed prices alongside lower scores; they do not show that the models are equivalent or that any price difference resulted from distillation.
Products Worth Considering
Logitech MX Brio Ultra HD 4K Webcam, Graphite
Apple iPad Air (4th Generation, 10.9-inch, Wi-Fi, 256GB) – Space Gray
LG UltraGear 27-Inch QHD OLED Gaming Monitor, 240Hz, 0.03ms, G-SYNC Compatible, FreeSync Premium Pro, HDMI 2.1
As an affiliate, we earn on qualifying purchases.
Earlier Concerns Over Model Access
The report follows previous disputes over attempts to obtain model outputs and reasoning. In February 2026, Anthropic said DeepSeek, Moonshot and MiniMax generated more than 16 million exchanges with Claude, according to the source material. The details of that earlier allegation are not included here, so the number should not be treated as proof of the current report’s claims.
In September, OpenAI said a “core cluster” of a July campaign to extract hidden reasoning from its models was linked to individuals associated with Moonshot AI. The source also says U.S. security agencies named both DeepSeek and Moonshot in an advisory. China’s commerce ministry called the accusations baseless. These are attributed claims and denials, not independent confirmation that the two labs used Astra outputs to train the models now named.
The source material says the U.S. government advisory described distillation as “a legitimate and useful technique in AI research.” That distinction is relevant: the method is widely used, while the current report concerns whether the teacher model belonged to a competitor and whether its provider’s terms or access controls were bypassed.
“Astra is “the most powerful model the company has released.””
— The Information, as quoted in its public opening
What the Public Record Lacks
The allegation remains unconfirmed publicly. The Information cited people with direct knowledge, but its full report is paywalled and the supplied account includes no documentary evidence, response from the named labs, or confirmation from OpenAI. Reuters reported that the companies did not immediately respond, which is not the same as a denial or confirmation.
Key details remain unknown: the number and type of Astra outputs allegedly used, the dates and scale of any activity, how the outputs were accessed, and how much they contributed to the released models. It is also unclear whether any use breached a contract or involved deception. The models’ launch statements describe other training methods but do not settle the question.
Responses and Evidence to Watch
The next developments to watch are direct responses from OpenAI, DeepSeek and Moonshot, along with any further reporting that supplies evidence about accounts, data or training pipelines. Public technical disclosures, contract enforcement actions or regulatory statements could clarify whether the alleged activity occurred and what rules may apply. Until then, the claim should be treated as a report, not an established finding.
For companies selecting AI services, the source recommends asking vendors about model and data provenance, testing systems on their own tasks, and maintaining a fallback option. It also advises users to avoid cheap, unofficial API resellers. The source says there is no immediate change to the consumer apps, and gives no indication that service access has been altered as a result of the report.
Key Questions
Which Chinese models are named in the report?
The report names Moonshot AI’s Kimi K3.1 Pro, released October 23, and DeepSeek-V4.1-Flash-1017, released October 17.
Has it been confirmed that the labs trained on GPT-6 Astra?
No. The Information reported the allegation, and Reuters summarized it, but the supplied public record contains no confirmation from OpenAI, DeepSeek or Moonshot and no disclosed training data.
What does distillation mean in this report?
Distillation is training a smaller model using answers from a larger “teacher” model. It is a standard technique; the dispute is whether a competitor’s outputs were used at scale in a way that violated terms or involved improper access.
Does the report prove that the models copied Astra?
No. The public account does not provide data volumes, methods or independent verification. It reports that the models benefited from Astra outputs, but the extent and circumstances remain unclear.
Source: ThorstenMeyerAI.com
