The popularity of Chinese AI models is growing faster than that of American ones

According to the OpenRouter platform, AI models processed a total of 27 trillion tokens in the week of March 30-April 5, 2026, which is a 19% increase compared to the previous week. Tokens are the basic units of text queries processed by the models: depending on the language and context, one token can correspond to a single character or a whole word.

The main contribution to this growth came from Chinese developments. Their aggregate figure reached 12.96 trillion tokens, an increase of more than 31% over the week. Instead, American models showed much more modest dynamics — only 3.03 trillion tokens with an increase of less than 1%. Chinese solutions have not only shown growth for five weeks in a row, but have been consistently outperforming the United States in this indicator.

The first six positions in the ranking were taken exclusively by Chinese models. The leader is Qwen3.6 Plus from Alibaba with 4.6 trillion tokens. The second place went to Xiaomi’s MiMo-V2-Pro — despite a 22% weekly decline, the model retained 3.08 trillion tokens. Qwen3.6 Plus Preview also held the third position with 1.64 trillion tokens.

The remaining places in the top 6 were taken by Step 3.5 Flash by StepFun, MiniMax M2.7, and DeepSeek V3.2. Thus, no American model made it to the top six in terms of the volume of tokens processed in the period.