China Overtakes US in Open Model Leadership in 2026, Leading in Downloads and Performance Metrics
Key point
In the 2026 open model ecosystem, China has seized leadership by overtaking the US in downloads, adoption rates, and performance metrics.
Details
As of 2026, China holds the lead in the open model ecosystem. Based on Hugging Face download data, Chinese models surpassed US models in July 2025, and as of September 2026, total downloads reached 3.2B, approximately twice that of the US. OpenRouter data also shows that the weekly throughput share of Chinese models has risen to over 80%.
Performance Gap and Market Share
According to the Artificial Analysis Intelligence Index, top Chinese models (such as GLM-5.3, score 45) score approximately 1.7 times higher than top US models (such as Inkling, score 26). Chinese labs maintain a technical lead by releasing models with scores equal to or higher than those achieved by US companies 2 to 6 months before the US companies reach those scores. In particular, in high-value industries such as software engineering, law, and finance, Chinese models have established themselves as alternatives to Claude and GPT, leveraging speed, low prices, and privacy protection.
Research and Adoption Status
The influence of Chinese models has surged in academic research as well. An arXiv analysis revealed that the mention rate of Chinese open-weight models exceeded 40%, surpassing US models (30%). Among these, the mention rate of the Qwen series was 30%, overtaking Llama (21%).
Regulation and Future Outlook
The US faces regulatory uncertainty amid the tension between the risk of misuse of open models and access restrictions. In contrast, China is emerging as a key tool for the global spread of AI services through rapid release speeds and ecosystem expansion.
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.