Chinese Open-Weight AI Models Lead US Models in Downloads and API Usage

In his written Congressional testimony, AI researcher Nathan Lambert revealed that Chinese open-weight AI models have reached 3.2 billion Hugging Face downloads compared to 1.6 billion for US models. Additionally, Chinese open models now generate over 80% of open-model traffic on the OpenRouter API platform. This trend highlights a major shift in global AI leadership, with Chinese open-weight models estimated to lag behind closed frontier models by only 2 to 5 months, compared to a 6 to 9 month gap for US open models. As a result, Chinese architectures like Qwen are increasingly becoming the default standard for open-source AI development worldwide. OpenRouter open-model traffic grew dramatically from 1 trillion tokens per week to nearly 80 trillion in a single year, driven heavily by Chinese model families like Qwen, GLM, and Kimi. The gap between open and closed frontier models is narrowest in agentic coding, where Chinese open weights are widely favored for tool-use and local execution.