DeepSeek, Moonshot Reportedly Trained Latest Models on GPT-6 Astra Outputs

What happened

The two models involved are Moonshot AI's Kimi K3.1 Pro, released last week, and DeepSeek's upgraded V4.1-Flash 1017, launched on Oct 17; The Information said both benefited from distilling Astra, citing people with direct knowledge, without disclosing the scale, methods or timeline of the distillation.

Distillation means training one company's model on the outputs of another company's model, and the report relayed by Reuters said the practice has become a flashpoint among frontier AI companies, which argue it lets rivals reproduce advanced capabilities more quickly and cheaply; Anthropic, OpenAI and Google have all made that argument. OpenAI said in a Sept 30 blog post that it had linked a core part of a distillation campaign that began in July to individuals associated with Moonshot AI, and Anthropic said earlier this month that new anti-distillation measures had cut the data harvested in distillation attacks on Claude by about 90% to 99%.

Key facts

1. Report: The Information reported on Oct 28, citing people with direct knowledge, relayed by Reuters on Oct 29.
2. Models involved: Moonshot AI's Kimi K3.1 Pro (released last week) and DeepSeek's V4.1-Flash 1017 (launched Oct 17).
3. Distilled source: outputs from OpenAI's GPT-6 Astra model.
4. Responses: OpenAI, DeepSeek and Moonshot AI did not immediately respond to Reuters' requests for comment.
5. Earlier disclosure: OpenAI said in a Sept 30 blog post that it had linked a core part of a distillation campaign that began in July to individuals associated with Moonshot AI.
6. Industry context: Anthropic said earlier this month that new anti-distillation measures had reduced the data harvested in distillation attacks on Claude by about 90% to 99%.