According to the latest calculations from the third-party data platform OpenRouter, the total weekly token usage of global AI large models reached 26.9 trillion tokens last week, showing a strong upward trend for four consecutive weeks. In this global competition of AI computing power and applications, Chinese AI large models have stood out, maintaining a high level of weekly usage volume.

Although the weekly token usage of American AI large models increased significantly compared to the previous week, the weekly usage volume of Chinese large models has already reached 1.81 times that of the United States. This means that Chinese large models have surpassed their American counterparts for three consecutive weeks, firmly holding the top position globally.

image.png

Tencent's model officially takes the lead, with price reduction strategy highlighting the advantage in inference cost

In the top three models ranked by usage volume, the first two are both Chinese AI large models. Among them, Tencent's Hy3 preview large model achieved a weekly token usage of 2.66 trillion tokens, successfully taking the lead on the list, seamlessly continuing after the free trial period, demonstrating high user retention.

The performance of the domestic rising star DeepSeek (Deep Seek) is also not to be underestimated, as its three large models remained on the list last week and their usage volume continued to rise. Among them, DeepSeek-V4-Flash achieved a weekly token usage of 2.06 trillion tokens, ranking second globally. The price reduction strategy is helping Chinese large models attract more global users.

Anonymous agent model emerges unexpectedly, high-performance base models are highly favored

Aside from the direct confrontation between Chinese and American giants, an anonymous large model called "Owl Alpha" appeared in last week's list and quickly climbed to the eighth position. Its weekly token usage doubled, becoming one of the most talked-about tech upstarts recently.

According to information from industry platforms, Owl Alpha is a high-performance base model designed specifically for Agent (intelligent entity) workflows. It not only supports complex tool calls and ultra-long context of millions of tokens, but also has excellent code generation and complex instruction execution capabilities.