China's LLM Weekly Token Usage Hits 23.45 Trillion, DeepSeek-V4-Flash Leads for Seventh Consecutive Week

According to reports from Sina Finance and other sources, the total weekly token consumption across China's large language model market reached 23.45 trillion, marking the tenth consecutive week of sustained high usage. The data indicates that Chinese AI model makers are seeing real-world application scale accelerate alongside their rapid model iteration.

DeepSeek-V4-Flash has been particularly dominant, holding the top spot for seven consecutive weeks. The model's efficient inference performance and competitive pricing have driven broad adoption among developers and enterprises.

Zhipu AI's GLM-5.2 is another notable riser, breaking into the top five for the first time. This reflects growing market recognition of Zhipu's continued investment in model capabilities. Improvements in multimodal and reasoning performance are likely key drivers behind GLM-5.2's usage growth.

Significantly, Chinese model vendors occupy all six top positions in the rankings. This suggests that domestic models have formed a clear concentration effect in the local market, with overseas models yet to establish competitive traction in Chinese-language scenarios and large-scale inference.

The 23.45 trillion weekly token figure represents massive real-world validation of China's AI application layer. It reflects not only growing developer trust in domestic models but also accelerating AI-native application penetration across industries.

From an infrastructure perspective, sustained usage growth places increasing demands on GPU cluster scheduling, inference engine optimization, and cost management — all critical variables supporting continued scale expansion.

Looking ahead, the key questions are whether DeepSeek-V4-Flash can extend its streak and whether GLM-5.2 will challenge the top three. As more vertical-scenario models launch, the competitive landscape could shift further.