Fueled by applications like AI Agents, the global consumption of tokens is experiencing a continuous upward trend. According to the latest data from OpenRouter, the total invocation volume of large AI models worldwide soared to 28.9 trillion tokens in the week from May 18 to May 24, marking a 7.4% increase from the previous week. This marks the fifth straight week of growth, underscoring a steady and sustained surge in demand for large model invocations.
Specifically, the weekly invocation volume of large AI models from China hit 9.223 trillion tokens, up by 19.89% week-over-week. In contrast, the volume for U.S. large AI models stood at 4.93 trillion tokens, reflecting a 16.27% increase from the prior week. For four consecutive weeks, Chinese large models have outpaced their U.S. counterparts in weekly invocation volume, securing the top spot globally and integrating more deeply into the global developer invocation ecosystem.
Among the top three models globally in terms of invocation volume, the first two are Chinese large AI models. DeepSeek-V4-Flash claimed the number one position, with a weekly invocation volume of 3.43 trillion tokens, representing a 66% week-over-week surge. Tencent's Hy3 preview came in second, with a weekly invocation volume of 3.07 trillion tokens, up by 16% from the previous week.