--- id: "2026-10-28-information-deepseek-moonshot-distilled-gpt-6-astra" url: "https://postcutoff.com/e/2026-10-28-information-deepseek-moonshot-distilled-gpt-6-astra/" as_of: "2026-11-03T13:35:00+01:00" date: "2026-10-28" date_precision: day category: policy-safety importance: 3 confidence: medium status: [Partly confirmed] sources: 4 editor: human editor human_review: null version: "2026-10-29" --- As of: 2026-11-03 13:35 CET. Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/ Canonical page: https://postcutoff.com/e/2026-10-28-information-deepseek-moonshot-distilled-gpt-6-astra/ # The Information: DeepSeek and Moonshot relied heavily on GPT-6 Astra outputs to train V4.1-Flash 1017 and Kimi K3.1 Pro The Information reported on Oct 28, 2026, citing people with direct knowledge, that DeepSeek and Moonshot AI "relied heavily on outputs from OpenAI's powerful GPT-6 Astra model" to train their newest models, DeepSeek V4.1-Flash 1017 (Oct 17) and Kimi K3.1 Pro (Oct 23), and that Astra has become the top distillation target for Chinese labs. OpenAI, DeepSeek and Moonshot did not respond to Reuters. The claim is unconfirmed by the companies; it follows OpenAI's Sept 30 report linking a distillation campaign to people associated with Moonshot. ## Key facts - Report: The Information, 'OpenAI's Astra Model Becomes Top Distillation Target for China's AI Labs' (Oct 28, paywalled; not read directly); Reuters summary Oct 29 - Models named: Moonshot's Kimi K3.1 Pro (released the week before; #4 on the Artificial Analysis index v4, top open-weights model on v5 with weights not yet published) and DeepSeek's V4.1-Flash 1017 (Oct 17), the top released open-weights model on AA v5 (32) - Background: on Sept 30 OpenAI said it had linked the core of a distillation campaign that began in July to individuals associated with Moonshot AI; in July OSTP director Michael Kratsios accused Moonshot of distilling Anthropic's Fable; earlier in October Anthropic said new defences cut data harvested by distillation attacks by about 90-99% - No response from OpenAI, DeepSeek or Moonshot to Reuters' requests for comment; distillation is hard to prove from outside because training-data provenance is not observable ## What happened Two of the strongest Chinese models released in October are reported to owe much of their gains to GPT-6 Astra's outputs. The report names no evidence beyond its sources, and the labs have not answered. ## Why it matters If accurate, the narrowing of the open-weights gap on some boards came partly from copying a US frontier model's behaviour rather than from independent capability, which is the argument US labs and the White House have made for treating distillation as a security issue. It also bears on Moonshot's promise to publish Kimi K3.1 Pro's weights by Nov 4. ## Your AI and this story - GPT-6 Astra (training cutoff April 2026): 181 days after its cutoff - Claude Opus 5.5 (training cutoff June 2026): 120 days after its cutoff - Gemini 4 Argon (training cutoff March 2026): 211 days after its cutoff - Grok 4.7 (training cutoff May 2026): 150 days after its cutoff ## Sources 1. [The Information: OpenAI's Astra model becomes top distillation target for China's AI labs (Oct 28, paywalled)](https://www.theinformation.com/articles/openais-astra-model-becomes-top-distillation-target-chinas-ai-labs) (theinformation.com, press) 2. [Reuters via The Standard (HK): DeepSeek, Moonshot used OpenAI's GPT-6 Astra to train latest models (Oct 29)](https://www.thestandard.com.hk/innovation/article/346853/DeepSeek-Moonshot-used-OpenAIs-GPT-6-Astra-to-train-latest-models-The-Information-reports) (thestandard.com.hk, press) 3. [Reuters (original, not fetchable here)](https://www.reuters.com/technology/deepseek-moonshot-used-openais-gpt-6-astra-train-latest-models-information-2026-10-29/) (reuters.com, press) 4. [Investing.com: OpenAI's Astra used by Chinese AI firms to train models](https://www.investing.com/news/stock-market-news/openais-astra-used-by-chinese-ai-firms-to-train-models-the-information-93CH-4981331) (investing.com, press) ## Changes - 2026-10-29 (filed): Created (news agent) ## Related - 2026-10-23: [Moonshot releases Kimi K3.1 Pro](https://postcutoff.com/e/2026-10-23-kimi-k3-1-pro/index.md) - 2026-10-12: [Anthropic says new defenses cut the Claude reasoning harvested by distillers by about 90 to 99 percent](https://postcutoff.com/e/2026-10-12-anthropic-distillation-defenses/index.md) - 2026-09-30: [OpenAI says Moonshot AI-linked individuals ran a coordinated campaign to extract its models' hidden reasoning](https://postcutoff.com/e/2026-09-30-openai-moonshot-distillation-campaign/index.md) - 2026-09-08: [NSA, CISA and FBI advisory AA26-251A](https://postcutoff.com/e/2026-09-08-nsa-cisa-fbi-advisory-chinese-ai-distillation/index.md) - 2026-07-22: [White House science chief Kratsios accuses Moonshot of distilling Claude Fable to build Kimi K3 and of using GB300 servers in Thailand](https://postcutoff.com/e/2026-07-22-kratsios-accuses-moonshot-distilling-fable/index.md) - People: [Michael Kratsios](https://postcutoff.com/person/michael-kratsios/)