Reuters: 300+ Documents Show China Distilling OpenAI Outputs for Qwen
TL;DR
- Reuters reviewed more than 300 Chinese papers, patents, corporate disclosures, recruitment notices and government documents showing distillation has moved into mainstream industrial use.
- A November 2024 Shanghai Jiao Tong paper selected about 5,000 answers from OpenAI's o1, translated them with GPT-4o mini, and built 10,750 bilingual examples to train Alibaba's Qwen2.5-72B.
- An eight-agency AI+Manufacturing policy package published in January includes guidance supporting model training, fine-tuning and distillation for industrial applications.
A Reuters review of more than 300 Chinese academic papers, patents, corporate disclosures, recruitment notices and government documents found that distillation of US frontier model outputs has moved beyond academic experiments into mainstream industrial use in China. The records name OpenAI's o1, GPT-4, GPT-4o and GPT-4o mini among the models whose outputs were used.
One example is specific enough to quote. In a November 2024 paper, Shanghai Jiao Tong University researchers described selecting about 5,000 answers from OpenAI's o1, translating them using GPT-4o mini and assembling 10,750 bilingual training examples to further train Alibaba's Qwen2.5-72B. An April 2025 Alibaba paper used GPT-4 and GPT-4o to generate training material in languages other than Chinese before training smaller Qwen models, and a February 2026 paper by Shanghai Jiao Tong and Meituan researchers used GPT-4o to generate plans for coordinating AI agents on programming problems.
The practice is not happening in a policy vacuum. A national AI+Manufacturing policy package published by eight Chinese government agencies in January includes guidance supporting model training, fine-tuning and distillation for industrial applications. Reuters notes the push is aimed at making AI cheaper and easier to deploy across China's manufacturing sector, with distillation described as unable to "reproduce a leading model's full capabilities or eliminate the need for advanced computing power."
The geopolitical read is sharp. A US official told Reuters that Washington believes "industrial-scale distillation is the main reason Chinese AI models have narrowed the gap with their American rivals," and that China's expanding use of distillation "could undermine the effectiveness of US restrictions on advanced chip sales to China." Two people familiar with Chinese developers' operations pushed back, telling Reuters distillation is only one factor alongside engineering, algorithms, data and model design.
Distillation is not illegal in itself, but OpenAI and Anthropic restrict the use of their models' outputs to develop competing systems. Anthropic has said it identified large-scale distillation of its models by DeepSeek, Moonshot AI and MiniMax. Shanghai Jiao Tong, Alibaba and Meituan did not respond to Reuters' requests for comment, and neither did the US Commerce Department or China's Cyberspace Administration. This is our newsroom's third China AI story today sitting on the same seam between chip controls and model access.
Originally reported by reuters.com
Read the original article →Original headline: Reuters: 300+ Documents Show Chinese Labs Distilling GPT-4o and o1 to Train Qwen Rivals