Recently, OpenRouter—a well-known AI model aggregation platform—released its latest weekly data, revealing that the global weekly usage of large AI models has surged to a staggering 164 trillion tokens.

Chinese models accounted for a remarkable 79.18 trillion tokens, representing 48% of the total. This means nearly half of the world’s AI processing volume is driven by Chinese large AI models.

In comparison, U.S. models contributed only 17.36 trillion tokens, roughly 10.6% of the total. This makes China’s large model usage 4.5 times that of the U.S.

Moreover, China has surpassed the U.S. in AI large model usage for 25 consecutive weeks. Currently, all top five entries on OpenRouter’s leaderboard are Chinese companies, with no U.S. firms making the cut.

Leading the pack is DeepSeek V4.1 Flash, followed by Zhipu GLM 5.3 Flash. In third place is Xiaomi MiMo-V2.6-Flash, fourth is Step 5 Preview from Step.cn, and fifth is Tencent’s Hy4 Preview.

Upon hearing this news, many netizens were excited, arguing that China’s AI large models have overtaken those of the U.S.—at least in terms of user adoption. After all, why would people choose Chinese models over American ones if they weren’t superior?

While this viewpoint isn’t entirely without merit, relying solely on OpenRouter’s data paints an incomplete picture.

After all, OpenRouter’s data comes from just one transit point and does not reflect the true global AI landscape. The platform primarily tracks developer and enterprise application traffic.

Many AI companies operate through their own channels, and a significant portion of usage goes unreported in OpenRouter’s statistics—think of applications like ChatGPT. For example, Google once announced that its AI large model, Gemini, processes 150 trillion tokens daily, far eclipsing OpenRouter’s figures. Thus, OpenRouter cannot be considered representative of the global total.

Secondly, another key factor behind China’s high large model usage is cost—many Chinese models are inexpensive or even free.

OpenRouter’s usage statistics include all these free interactions, which inflate token counts without generating meaningful revenue.

Additionally, when comparing pricing, some domestic AI models cost 90% less than their U.S. counterparts. This affordability encourages developers to experiment with Chinese AI large models, especially since they can test free tiers before committing to low-cost paid options.

Nevertheless, there’s no denying that China’s AI is growing increasingly powerful. Whether free or paid, the sheer volume of users serves as the best evidence of its rising influence.