The latest data from the online AI hosting platform OpenRouter shows that the global AI market competition landscape has undergone a major reversal in February this year. Chinese open-source models represented by MiniMax, Moonshot AI, Zhipu, and DeepSeek have, for the first time, comprehensively surpassed American models in global Token usage.

image.png

As an important indicator of global AI development activity, OpenRouter aggregates various large models from Google, Anthropic, and leading Chinese companies. Data shows that during the period from February 9 to 15, the weekly call volume of Chinese models reached 4.12 trillion Tokens, surpassing American models' 2.94 trillion Tokens for the first time. In the following week, the lead of Chinese models further expanded, with the weekly call volume reaching as high as 5.16 trillion Tokens, a 127% increase within three weeks, while the usage of American models showed a downward trend during the same period.

In the specific rankings for February, Chinese large models performed exceptionally well:

  • Champion: The MiniMax M2.5, which was launched just two weeks ago, won with a call volume of 4.55 trillion Tokens. Its programming and search capabilities are now sufficient to compete with the main products of OpenAI and Anthropic.

  • Runner-up: Moonshot AI's Kimi K2.5 came second with 4.02 trillion Tokens.

  • Top five positions: Chinese companies occupied three spots (MiniMax, Moonshot AI, DeepSeek), collectively capturing nearly two-thirds of the traffic share in the top five.

Analysts point out that this explosive growth stems from the concentrated efforts of domestic manufacturers before and after the Spring Festival. The launch of models such as Kimi K2.5 is believed to have narrowed the technological gap between China and the U.S. in AI research and development to the lowest level in history. At the same time, the enterprise market in China also experienced a surge, with the daily Token usage of Chinese enterprises increasing by 263% in the second half of 2025 compared to the first half, showing the profound potential of AI application and implementation.