China's AI gains reflect real innovation, not just distillation, Cohere CEO says

September 18, 2026 at 06:22 PM2 min

This digest was compiled by AI from multiple sources — links to the originals are below.

Cohere CEO Aidan Gomez said Chinese AI models are 'world class' and that the U.S. lead is 'evaporating very quickly.' He argued that China's progress cannot be explained by distillation alone, as some Chinese models now beat the best American models on certain benchmarks. His comments run counter to claims by U.S. labs like Anthropic that Chinese labs have used distillation amounting to theft.

Key Facts

Cohere CEO Aidan Gomez said Chinese AI models are 'world class' and the U.S. lead is 'evaporating very quickly.'
Gomez said some Chinese models beat the best American models on certain benchmarks, which cannot be achieved through distillation alone.
Anthropic has repeatedly said Chinese AI labs used distillation, calling it theft.
A Morgan Stanley survey found 80% of respondents in China use AI for personal purposes at least once a week, compared with 54% in the U.S.

Distillation Debate

U.S. companies and Washington have accused Chinese labs of using distillation to gain ground, denouncing their efforts as theft. Aidan Gomez, CEO of Cohere and co-author of the 2017 paper 'Attention Is All You Need,' said distillation was 'definitely true to an extent' but cannot explain all of China's progress. Gomez noted that Chinese models now outperform the best American models on some benchmarks, which he said is impossible through copying or distillation alone. Anthropic, the creator of Claude, has on numerous occasions said Chinese AI labs used distillation, calling it theft.

AI Adoption Gap

A Morgan Stanley survey found that 80% of respondents in China use AI for personal purposes at least once a week, compared with 54% in the U.S. The survey indicates China is ahead in the adoption of AI models among the general population, even as U.S. labs lead in frontier model development. Morgan Stanley's figures are based on survey respondents and may not exactly match adoption by the general population.