DeepSeek and Moonshot AI relied heavily on outputs from OpenAI’s GPT-6 Astra to train their latest models, The Information reported on Wednesday, according to Reuters. Why it matters: if the reporting is right, two of China’s best-known AI labs built their newest releases partly on a US rival’s model outputs — the practice known as distillation, which has become one of the sharpest flashpoints in the US-China AI race. None of the three companies has confirmed it.
October 29, 2026 — Tested Intelligence Review Team
What was reported
Reuters, citing The Information’s reporting and people with direct knowledge of the matter, said the distillation benefited two specific releases: Moonshot AI’s Kimi K3.1 Pro, which was released last week, and DeepSeek’s upgraded V4.1-Flash 1017 model, which launched on October 17. Reuters noted that OpenAI, DeepSeek, and Moonshot AI did not immediately respond to its requests for comment, so the claim currently rests on The Information’s sourcing and has not been confirmed or denied by any of the companies involved.
What distillation is, and why labs care
Distillation means training a smaller or cheaper model on the outputs of a larger, more capable one, so the student model learns to mimic the teacher’s reasoning and answers. Done at scale, it can reproduce advanced capabilities far more quickly and cheaply than training from scratch — which is exactly why frontier labs argue it amounts to copying. OpenAI, Anthropic, and Google have all argued publicly that the practice lets rivals close the capability gap at a fraction of the original cost.
This did not come out of nowhere
In a September 30 blog post, OpenAI said it had linked a core part of a distillation campaign that began in July to individuals associated with Moonshot AI, and it disrupted that activity. Earlier this month, Anthropic said new measures designed to stop rivals from copying Claude’s capabilities had cut the data harvested in distillation attacks by about 90% to 99%. Wednesday’s report is the first to tie the alleged Astra distillation directly to two named, already-shipped models — Kimi K3.1 Pro and DeepSeek’s V4.1-Flash 1017.
Why it matters
For readers, the honest state of play is this: the allegation is serious, specifically sourced, and unconfirmed. If it holds up, it strengthens the case US labs are making for tighter controls on model access and adds pressure on Chinese labs whose recent models have been praised for closing the gap at much lower prices — the obvious question being how much of that price advantage came from standing on a rival’s outputs. If it does not hold up, it will be a reminder of how hard distillation is to prove from the outside. Either way, expect the response to land in access rules, verification tiers, and export policy rather than in press statements. Everything AI, tested — and increasingly, everything AI, checked for whose work it was trained on.
Frequently asked questions
What did The Information report about DeepSeek and Moonshot?
As reported by Reuters, The Information said both labs relied heavily on outputs from OpenAI’s GPT-6 Astra to train their latest models — Moonshot’s Kimi K3.1 Pro and DeepSeek’s V4.1-Flash 1017.
What is distillation in AI?
Distillation is training one model on another model’s outputs so it learns to mimic the stronger model’s behavior, which can reproduce advanced capabilities more quickly and cheaply than training from scratch.
Have OpenAI, DeepSeek, or Moonshot confirmed the report?
No. Reuters reported that none of the three companies immediately responded to requests for comment.
Has OpenAI raised distillation concerns before?
Yes. In a September 30 blog post, OpenAI said it had linked part of a distillation campaign that began in July to individuals associated with Moonshot AI.
Sources: Reuters (October 29, 2026), reporting on The Information; OpenAI blog post of September 30, 2026, as cited by Reuters.