In the race for computing power in the AI era, China's large models are achieving a "surpassing turn" at an astonishing speed. According to the latest monitoring data from OpenRouter, during the period from March 2 to March 8, the weekly call volume of China's large models reached 4.19 trillion Tokens, surging by 34.9% compared to the previous week.

This is not just a simple numerical breakthrough, but also the second consecutive week that China's large models have surpassed those in the United States in terms of call volume (the call volume of US large models during the same period was 3.63 trillion Tokens, a decrease of 8.5% compared to the previous week). This "rise and fall" reflects the unprecedented boom in China's AI application ecosystem.

Three Chinese companies among the global top five, MiniMax remains the leader

In the global ranking of large model call volumes, Chinese products have taken three spots:

  • MiniMax M2.5: Maintains the number one position globally with a weekly call volume of 1.87 trillion Tokens, increasing by 15% compared to the previous week. With its high cost-effectiveness and stable output, MiniMax has become the preferred choice for developers.

  • DeepSeek V3.2: Ranked third globally with a weekly call volume of 0.83 trillion Tokens, demonstrating its deep expertise in handling complex tasks.

  • Stepwise Star Step3.5 Flash: The most impressive newcomer this week, with a weekly call volume of 0.75 trillion Tokens, surging by 69% compared to the previous week, successfully entering the global top five.

Notably, Kimi K2.5, which had been performing strongly, dropped to sixth place globally this week, but its call volume still maintained a steady growth of 8%.

A dual victory in computing power and application

The surge in call volume of China's large models is no accident. On one hand, manufacturers such as MiniMax and DeepSeek have continuously optimized their technology, significantly reducing the cost per Token, making it feasible and effective to use them. On the other hand, China's rich terminal customer scenarios and industry application demands provide a vast battlefield for practical testing of large models.

As the global AI competition enters its second half, it's no longer just about the scale of parameters, but who can integrate more deeply into the production process. The continuous surpassing of the US by China's large models may indicate a profound shift in the center of gravity of the global AI industry. In the struggle between "excess computing power" and "application hunger," China's AI is accelerating on its own path.