Chinese labs now dominate open-weight AI, Lambert briefs Congress

The top three Chinese open-weight models score 45, 42 and 44 on the Artificial Analysis Intelligence Index. The leading American open models score 26 and 23.

Those are the numbers Nathan Lambert put in front of a group of Congressional members and staff this month, and published as prepared remarks on Interconnects. The Chinese leaders are Z.ai's GLM-5.3 and GLM-5.3-Flash, and Moonshot AI's Kimi K3. The U.S. entrants are Thinking Machines' Inkling and Inkling Small, and Nvidia's Nemotron 3 Ultra.

"Since about April 2025, Chinese AI companies have been the clear leader in open-weight models," Lambert writes.

The usage data tracks the benchmark data. Chinese models "have grown from ~70% market share to over 80% of usage," and open-model traffic through OpenRouter has itself grown from roughly 1T tokens processed per week in September 2025 to about 80T tokens per week today. On Hugging Face, "China's download lead has grown to about 1.6B – with a total of 3.2B downloads, twice that of America's total."

Academic research shows the same pattern. Llama peaked at about 23% of AI/ML papers in fall 2024 and now sits at 21%, while Qwen has climbed from 7.5% to 30%. "Overall, any Chinese open weight model is mentioned in over 40% of papers, over the U.S.'s 30%, with China's share continuing to grow," Lambert writes.

The customer list is American. "Many prominent technology companies and startups have been building on Chinese open-weight models for their AI features, such as Harvey, the legal agent, Cursor, the coding agent, and DoorDash's use of Kimi models, Airbnb's use of Qwen, or Perplexity's use of DeepSeek."

Lambert's most striking claim is what does not close the gap. He estimates that fully preventing distillation, for instance via know-your-customer tools at Anthropic and OpenAI, would only increase the American open-weight lead by 1-2 months. The rest of the gap, he argues, is Chinese labs shipping capability faster: Chinese open weights trail the closed American frontier by 2-5 months, while American open weights trail it by 6-9 months.

"This represents a substantial source of soft power, influence, and potential for the organizations that enable this broad access to transformative intelligence," Lambert writes.