Chinese AI developers DeepSeek and Moonshot AI relied heavily on outputs from OpenAI’s GPT-6 Astra model to train their latest models, according to The Information, which reported it on 28 October 2026 citing people with direct knowledge of the matter, as relayed by Reuters.
The report names two models: Moonshot’s Kimi K3.1 Pro and DeepSeek’s V4.1-Flash 1017. Reuters said OpenAI, DeepSeek and Moonshot AI did not immediately respond to its requests for comment.
Business Pill · Data Flywheel
An evergreen Business Engineer explainer on data flywheels. It explains the general idea only and does not discuss this report.
The key insight: By our arithmetic, GPT-6 Astra’s list price for output tokens is 5 times Kimi K3.1 Pro’s and about 42 times DeepSeek V4.1 Flash’s; The Information reports that both of those cheaper models leaned on Astra’s outputs in training.
Reuters wrote that “Moonshot AI’s Kimi K3.1 Pro model, which was released last week, and DeepSeek’s upgraded V4.1-Flash 1017 model, which was launched on October 17, benefited from distilling Astra, The Information said.”
The Information’s own headline calls Astra the “Top Distillation Target for China’s AI Labs”. Its article is behind a paywall; we read Reuters’ account of it and the headline, and did not see the people it cites.
Reuters described distillation as a practice that “has become a flashpoint for leading AI companies as US-China technology competition intensifies”, and said firms including Anthropic, OpenAI and Google have argued it lets rivals reproduce advanced capabilities more quickly and at lower cost.
List price per million output tokens on each company’s API pricing page, read 29 October 2026. Sources: OpenAI, Moonshot (Kimi) and DeepSeek pricing pages. Ratios are our arithmetic.
What Was Posted on X
Chinese labs moved their distillation efforts from Claude to OpenAIโs GPT-6 Astra
per The Information
Anthropic said this month that tighter limits, account bans and proxy-network blocks cut the data distillers harvest from Claude by 90% to 99%.
Rohan Paul’s post, citing The Information, says Chinese labs moved their distillation efforts from Claude to OpenAI’s GPT-6 Astra, and notes Anthropic said tighter limits, account bans and proxy-network blocks cut the data distillers harvest from Claude by 90% to 99%.
SITUATION EXPLAINED: Anthropic cut off Chinese labs from distilling Claude, so they reportedly switched to OpenAI’s GPT-6 Astra.
โข Moonshot’s Kimi K3.1 Pro and DeepSeek’s V4.1-Flash 1017, both released this month, were reportedly trained on Astra’s outputs, and neither lab has commented โข Distillation means training a smaller “student” model to copy a stronger “teacher” model’s answers, and it works through a normal API without the teacher’s weights โข In February, Anthropic said DeepSeek, Moonshot and MiniMax used about 24,000 fake accounts to run over 16 million chats with Claude โข Distillers mostly want Claude’s step-by-step reasoning, and in Anthropic’s own test, a model trained on it scored 80.4% on a coding benchmark vs 71.8% when trained on summaries โข Anthropic now hides that reasoning from distillers and bans their accounts, often within 2 hours of a campaign starting โข Anthropic reportedly told investors distillation made up 5% to 7% of its revenue, and it’s walking away from that money to shut it down โข OpenAI said in September that it traced a distillation campaign that began in July to people linked to Moonshot
@theojaffee: “It’s not all distillation. I think there’s a little bit of cope going on here. They genuinely have amazing talent. They’re genuinely able to work very well with very limited resources compared to what US labs have in terms of compute.”
MTS on X, 29 Oct: Anthropic cut off Chinese labs from distilling Claude, so they reportedly switched to OpenAI’s GPT-6 Astra; in February Anthropic said DeepSeek, Moonshot and MiniMax used about 24,000 fake accounts to run over 16 million chats with Claude, and Anthropic reportedly told investors distillation made up 5% to 7% of its revenue.
What OpenAI Said Before
On 30 September, OpenAI said it had disrupted “a coordinated campaign designed to extract protected reasoning from our models”, with the earliest activity in the first week of July. It defined adversarial distillation as “the systematic and unauthorized use of one modelโs outputs or reasoning to help train, reproduce, or improve another model”.
OpenAI said the activity spiked on 24 and 25 July with 16,000 requests from over 4,000 users, and that it found related activity across “a cluster of more than 15,000 users, which we fully disrupted by July 28”. It wrote: “we attribute a core cluster of the activity to individuals associated with Moonshot AI, the developer of Kimi.”
OpenAI also said “It is unclear whether all operators we observed during the relevant time period originated from a single actor.” Its September post did not name DeepSeek.
What Anthropic Reported
Anthropic said in a post on its own defences that new methods have “reduced the amount of data harvested by distillers by around 90 to 99%”. It described distillation attacks as “covert, industrial-scale campaigns to extract the capabilities of one model and replicate them in another”.
The Price Gap
OpenAI’s API pricing page lists gpt-6-astra at $10.00 per million input tokens and $50.00 per million output tokens. Moonshot’s pricing page lists kimi-k3.1-pro at $2.00 and $10.00. DeepSeek’s pricing page lists DeepSeek-V4.1-Flash-1017 at $0.3 per million input tokens on a cache miss and $1.2 per million output tokens, both at peak rates.
By our arithmetic, Astra’s list price for output tokens is 5 times Kimi K3.1 Pro’s and about 42 times DeepSeek’s V4.1 Flash. These are list prices on the companies’ pages, not a measure of what any modelcost to train.
The Structural Read
As we read the report, the claim is narrower than a general accusation: Reuters says The Information named two specific models, Kimi K3.1 Pro and V4.1-Flash 1017, and said both “benefited from distilling Astra”.
OpenAI’s own September post describes distillation as “the systematic and unauthorized use of one modelโs outputs or reasoning to help train, reproduce, or improve another model”.
Anthropic says its new methods have “reduced the amount of data harvested by distillers by around 90 to 99%”; OpenAI’s post lists account enforcement, technical controls and partner coordination as its response.
OpenAI, 30 September 2026
“we attribute a core cluster of the activity to individuals associated with Moonshot AI, the developer of Kimi.”
Three Implications
FOR OPENAIOpenAI’s September post says it “banned or restricted fraudulent accounts, strengthened signup and infrastructure controls, and expanded monitoring for related networks”.
FOR BUYERS OF MODEL APIS Each company’s pricing page lists Astra at $50.00 per million output tokens, Kimi K3.1 Pro at $10.00 and DeepSeek V4.1-Flash-1017 at $1.2 at peak.
FOR THE LABS NAMED Reuters said OpenAI, DeepSeek and Moonshot AI did not immediately respond to its requests for comment.
The framework argues: “The transformer is a commodity now; what changes between paradigms is where intelligence accumulates and who can capture it.”
The framework does not discuss this report or distillation between labs.
What Is Not Established
The claim that the two models were trained on Astra outputs rests on The Information’s sources as reported by Reuters. OpenAI, DeepSeek and Moonshot AI had not responded to Reuters when it published, and none of the three has confirmed or denied the report in a statement we could find.
The report, as relayed, does not give the volume of Astra outputs involved or say how the companies are said to have obtained them. OpenAI’s September attribution covered activity from July and did not name the models in this report.
This Week In The AI Supercycle
Execution Fault Lines, Concentrated Demand, and Geopolitical Firewalls
This week’s issue: the AI build is not short of demand for its parts. It is short of cheap money, and its end revenue rests on a narrow base of large, often venture-funded buyers.
The Information reports that DeepSeek and Moonshot relied heavily on outputs from OpenAI’s GPT-6 Astra to train Kimi K3.1 Pro and DeepSeek V4.1-Flash 1017. OpenAI said in September that it attributed a core cluster of a distillation campaign to individuals associated with Moonshot AI; the companies had not commented on this report when Reuters published.
A note on sourcing. The training claim comes from The Information, as relayed by Reuters on 28 October 2026; The Information’s article is paywalled and we read only its headline. OpenAI’s and Anthropic’s statements come from their own posts. Prices come from each company’s API pricing page, read on 29 October 2026; ratios are our arithmetic.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
