Chinese AI models have surpassed their American counterparts in developer usage on platforms like OpenRouter and Hugging Face as of mid-2026. These models now process over three times the weekly token volume compared to U.S.-based systems, marking a significant shift since February when this trend began.
One major factor driving this shift is cost. Developers facing rising fees from providers like OpenAI and Anthropic are turning to Chinese alternatives such as DeepSeek, GLM 5.2, and Kimi K3. These options offer performance comparable to U.S. models but at a fraction of the price. For example, Zhipu's GLM 5.2 model charges roughly $1.40 per million input tokens and $4.40 per million output tokens, while Anthropic's Opus 4.8 comes in at $5 and $25 respectively.
Rising Popularity of Open-Weight Models
OpenRouter, which manages API calls across multiple AI systems, reports Chinese open-weight models handle more than triple the tokens of American models each week. Unlike closed models sold per token by U.S. companies, many Chinese labs release their model weights openly, enabling developers to run them without ongoing fees.
During the week of February 9 to 15, Chinese AI systems processed 4.12 trillion tokens on OpenRouter, surpassing the 2.94 trillion tokens managed by American systems. This marked the first time U.S. models lost their dominant position, after holding close to 70% of usage through most of 2025.
Downloads also reflect this trend. Chinese models accounted for 41% of Hugging Face downloads in spring, overtaking U.S. releases. Vercel’s Production Index showed open-weight models were behind 29% of all tokens handled on its AI gateway in June 2026, nearly tripling their share from April.
The lower cost and open-weight nature of these models make them increasingly attractive for startups and enterprises looking to optimize AI usage economically. This shift signals a new competitive landscape in AI development, with Chinese labs gaining ground fast.
