China's lead in open-weight AI

Chinese open-weight AI models lead American counterparts on capability benchmarks, download volume, and academic citations as of 2026. A briefing prepared for U.S. Congressional members states that since about April 2025, Chinese AI companies have been the clear leader in open-weight models. On the Artificial Analysis Intelligence Index, the briefing finds the top American open models rank below 15 Chinese models, with Chinese leaders scoring 42-45 versus American leaders at 23-26. China took the Hugging Face download lead in July 2025 and holds roughly a 1.6 billion download advantage. Alibaba's Qwen ranked as the most downloaded model series on Hugging Face in both 2025 and 2026, and a MIT study confirmed Chinese open-source models lead US models in total downloads overall. On academic adoption, the briefing states Chinese open-weight models appear in 38% of AI papers versus the US at 28%, with Qwen cited in 30% of papers compared to Meta's Llama at 21%. The briefing estimates distillation from American frontier models accounts for only about one to two months of the performance gap, placing the bulk of the Chinese lead outside that technique. The briefing also notes that open-weight models are reaching capability thresholds that enable new cybersecurity risks, while researchers rely on them partly because closed models decline to assist with defensive security and biology research.