Chinese firms scale AI distillation using U.S. model outputs
Reuters reports that distillation has moved from academic research into mainstream industrial use in China, based on its review of more than 300 Chinese documents. The practice uses outputs from a powerful “teacher” model to train a smaller “student” model.

Key takeaways · 4
- 01
Distillation is being used beyond academic research, with Chinese firms developing automated workflows for it.
- 02
Shanghai Jiao Tong researchers described using about 5,000 OpenAI o1 answers to assemble 10,750 bilingual examples for Alibaba’s Qwen2.5-72B.
- 03
The technique can retain selected skills in smaller systems, but does not recreate a leading model’s full capabilities.
- 04
OpenAI’s terms restrict using its outputs to develop competing models, subject to exceptions.
From research to industrial workflows
Reuters reported that its review of more than 300 Chinese documents found distillation had spread from academic research into mainstream industrial use.[1] Chinese companies are building software to automate the process, while specialist workers prepare the training material.[1] The basic approach is to use responses from a powerful teacher model to train a smaller student model.[1] An American model can provide a difficult answer once; a team can then retain useful examples and train a local system around them.[1] This offers a way to transfer selected capabilities without repeatedly relying on the larger model for each task.[1]
Examples from Chinese researchers
Shanghai Jiao Tong University researchers described selecting about 5,000 answers from OpenAI’s o1 model, translating them with GPT-4o mini, and assembling 10,750 bilingual training examples for Alibaba’s Qwen2.5-72B model.[1] In other work, researchers used GPT-4 and GPT-4o to generate training material in languages other than Chinese, then refined and scored responses before training smaller Qwen models.[1] Shanghai Jiao Tong University and Meituan researchers also described using GPT-4o to generate plans for AI agents tackling programming problems, then training a smaller model to organize those tasks.[1]
Capability gains have limits
Reuters said distillation can help developers preserve selected skills in systems that are smaller and cheaper to operate.[1] It cannot reproduce a leading model’s full capabilities or eliminate the need for advanced computing power, however.[1] Washington believes industrial-scale distillation is the main reason Chinese AI models have narrowed the gap with American rivals, but Reuters reported that distillation was one factor among advances in engineering, algorithms, data and model design.[1]
Policy, chips and model terms
Beijing is encouraging distillation to make AI cheaper and easier to deploy across China’s manufacturing sector, and a national AI+Manufacturing policy package published by eight government agencies in January supported model training, fine-tuning and distillation for industrial applications.[1] Shanghai and Shenzhen have also encouraged techniques intended to reduce computing requirements.[1] Washington said on Nov. 16 that it would further restrict China’s access to chips from companies such as Nvidia.[1] Reuters reported that U.S. officials raised concerns with Chinese regulators about collecting outputs from leading American models, while Chinese officials defended distillation as a standard technique and made no concrete commitments.[1] OpenAI’s terms restrict use of its output to develop competing models, subject to exceptions; Anthropic restricts training competing systems without permission.[1]
For teams weighing smaller, lower-cost models, distillation offers a way to target selected skills, not a substitute for a leading model’s full capabilities. The reported examples also make model-provider terms and the continuing need for advanced computing part of the deployment decision.
Why it matters
AI training for your team
kju.ai is built for teams: daily sessions, team analytics, and rollout support for your organization.
How this developed
17 November 2026
Chinese firms scale AI distillation using U.S. model outputs