Export controls appear to be working: US AI labs pulling ahead of Chinese counterparts on capability benchmarks
Key Points
- US AI labs are widening their capability lead over Chinese competitors, with a steeper trajectory across recent model releases from OpenAI, Anthropic, and others versus DeepSeek, Alibaba, and Kimi.
- US export controls on chips appear to be producing measurable effects on the pace of Chinese AI model development.
- The sustainability of this gap depends on whether export restrictions remain in place and whether Chinese labs can develop workarounds.
Summary
US AI Labs Pulling Ahead on Benchmarks; Export Controls Appear to Be Working
The capability gap between US and Chinese AI labs is widening, according to benchmark comparisons discussed in this segment. Looking across major model releases—GPT-4o, OpenAI o1, o3, Claude Opus 4, GPT-5, and GPT-5.5 from US labs versus DeepSeek r1, v3, v4 pro, Alibaba's Qwen, and Kimi from Chinese competitors—the US trajectory appears steeper.
DeepSeek's v4 pro represents a notable jump in Chinese capability, but the overall trend favors American labs. The observation here hinges on export controls: chip restrictions appear to be producing measurable effects on relative model development velocity.
The caveat is timing. Whether this gap persists depends on whether export controls remain in place and whether Chinese labs find workarounds. For now, the empirical signal suggests the restrictions are having their intended effect.
Every deal, every interview. 5 minutes.
TBPN Digest delivers summaries of the latest fundraises, interviews and tech news from TBPN, every weekday.