AIThis post was created with the assistance of artificial intelligence (AI).

🔍 Read the full analysis: Why The GPT-6 Astra Question Matters For China’s New AI Models on ThorstenMeyerAI.com

Buying for a business?Offer from Amazon

Get business pricing on office and shipping supplies

  • Business-only prices and quantity discounts
  • Tax-exempt purchasing
  • Multiple users, one account, clear invoices
As an affiliate, we earn on qualifying purchases.

TL;DR

The Information reported on Oct. 28 that DeepSeek and Moonshot AI relied heavily on OpenAI’s GPT-6 Astra outputs to train recent models. Reuters said the companies did not immediately respond, and the public reporting provides no figures or details on how the outputs were obtained. The allegation matters because it puts model training practices, access controls and contractual rules under scrutiny, but it is not independently established by the information available here.

The Information reported Oct. 28 that China-based AI labs DeepSeek and Moonshot AI relied heavily on outputs from OpenAI’s GPT-6 Astra to train models released this month. Reuters summarized the allegation, naming Moonshot’s Kimi K3.1 Pro and DeepSeek’s V4.1 Flash update; the public record described in the reports does not establish how much Astra output was used or how it was obtained.

The Information’s report, by Jing Yang, Juro Osawa and Qianer Liu, cited people with direct knowledge, according to Reuters. Its public-facing summary described Chinese AI labs as a growing customer base seeking to extract Astra’s capabilities. The article’s full reporting is paywalled, limiting what can be checked from its publicly available text.

Reuters said the models identified were Kimi K3.1 Pro, released Oct. 23, and DeepSeek-V4.1-Flash-1017, launched Oct. 17. It reported that neither OpenAI nor the two Chinese labs immediately responded to requests for comment. The supplied reporting contains no response from the companies confirming or denying the allegation.

The reports do not provide usage totals, training methods or account details, leaving open questions about Astra’s model economics. Moonshot’s launch post attributed Kimi’s gains to more post-training compute and large-scale reinforcement learning. DeepSeek’s update note, as described in the source material, did not discuss training data. Those statements do not resolve whether Astra outputs were used.

At a glance
reportWhen: Reported Oct. 28, 2026; follow-up cover…
The developmentThe Information reported that DeepSeek and Moonshot AI used GPT-6 Astra outputs to train recent models, prompting renewed scrutiny of AI distillation and access rules.
Thorsten Meyer AI
Explained

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.

Step 1 · Ask

Send the teacher prompts

A strong “teacher” model gets a huge number of carefully chosen tasks.

Step 2 · Collect

Keep answers and reasoning

The step-by-step working is worth far more than the final answer alone.

Step 3 · Train

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).

  • Trained on full reasoningAnthropic experiment80.4%
  • Trained on summarized reasoningAnthropic experiment71.8%

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 Astra Outputs Matter

AI distillation is a way to train a smaller model to imitate answers from a larger one. It can reduce the cost of developing and serving models, and it is a standard research technique when a lab trains from its own systems or otherwise has permission. The controversy here is not the technique alone, but the allegation that one company’s model outputs may have been used to train competing products at scale.

OpenAI’s terms, according to the source material, prohibit using its outputs to build competing models. If the reported conduct is substantiated, the dispute could involve contract compliance and, depending on how access was obtained, possible misuse of accounts. Those are questions requiring evidence and legal analysis; the current reporting does not establish a violation or its consequences.

For AI providers, the episode highlights the difficulty of controlling how model outputs are reused once customers can query a system. For businesses choosing models, it makes training-data provenance and vendor due diligence more relevant. For users, the reporting does not establish that the named products are unsafe or that their current behavior will change.

Products Worth Considering

As an affiliate, we earn on qualifying purchases.

Earlier Scrutiny of Model Access

The allegation follows earlier disputes over attempts to extract capabilities from rival models. The source material says OpenAI had previously linked a July campaign targeting hidden reasoning to people associated with Moonshot AI, and that U.S. security agencies named DeepSeek and Moonshot in an advisory. These points are part of the reported backdrop; they do not independently prove the new claim about Astra training.

The same material says an advisory described distillation as a legitimate and useful AI research technique, while warning about problematic extraction. It also reports that China’s commerce ministry called accusations against Chinese AI firms baseless. These competing positions underscore the need to distinguish the method itself from alleged unauthorized access or use.

The source cites different model prices and scores to illustrate commercial stakes: 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 gives scores of 57, 46 and 32, respectively, on the same independent index, but does not identify the index in the supplied text. These figures do not show that the models are equivalent or establish how any model was trained.

“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 Claim

The central allegation remains unconfirmed in the available public record. The Information cited people with direct knowledge, but its full report is paywalled, and the supplied material includes no documents, technical analysis or named sources that readers can independently examine. Reuters reported that the three companies had not immediately responded to its requests.

It is not clear how many Astra outputs were involved, whether they included hidden reasoning, how the outputs were collected, or whether the labs used direct accounts, resellers or another route. The supplied material also does not establish whether the reported activity violated a specific contract or law. Claims about prior extraction efforts and security advisories provide context, not proof of this particular allegation.

Company Responses and Further Evidence

The next meaningful developments would be responses from OpenAI, DeepSeek and Moonshot, along with any additional reporting that specifies the scale and method of the alleged training. Technical evidence, account records or regulatory findings could help clarify whether Astra outputs were used and under what conditions.

Until then, companies evaluating AI suppliers can ask vendors about training-data provenance, access controls and applicable terms, and test models on their own workloads. The source material also advises users to avoid cheap, unofficial API resellers. The report does not say that ordinary users need to change how they use the named models today.

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, citing people with direct knowledge. The public information provided here does not include the full paywalled account or supporting evidence.

Which models were named?

The reports identified Moonshot’s Kimi K3.1 Pro, released Oct. 23, and DeepSeek-V4.1-Flash-1017, launched Oct. 17.

Is distillation itself improper?

No. Distillation is a common technique for training a smaller model from a larger model’s outputs. The dispute concerns whether a rival’s outputs were used at scale and whether access or use breached the provider’s terms.

Has OpenAI confirmed that its terms were violated?

The supplied reporting does not establish a violation. It says OpenAI’s terms prohibit using its outputs to build competing models, but the public record lacks details about the alleged activity, access method and relevant contract circumstances.

What should users or businesses do now?

The reporting does not indicate an immediate change for users of the apps. Businesses can ask AI vendors about data provenance and access controls, while users should be cautious about unofficial API resellers, as advised in the source material.

Source: ThorstenMeyerAI.com

This content is for general information only and is not financial, tax or legal advice. Consult a qualified professional for decisions about your money.
You May Also Like

AI Plans May Be Shifting: What Grok Costs And OpenAI’s Rules Could Mean

Elon Musk cited Grok Bot’s costs, while OpenAI set account-sharing limits and a shared ChatGPT-Codex allowance. Key details remain unconfirmed.

Trending Lifestyle Hacks For Tying Gym Shorts And Drawstrings

New trending methods for tying gym shorts and drawstrings are gaining popularity, offering improved comfort and style. Here’s what you need to know.

Capgemini: Multi-agent AI Moves Into Banking And Insurance, But Orchestration Remains Limited

Capgemini research says 59% of banks and insurers are piloting or deploying multi-agent AI, but only 10% have enterprise-wide orchestration.

Mattel Surges In Global Coverage

Search interest and media coverage of Mattel have surged significantly, with 25 mentions in recent window, indicating rising global attention without a clear trigger.