Logo of DeepSeek Photo: VCG

DeepSeek's large language model (LLM) V4.1-Flash-1017 recorded 33.6 trillion tokens in weekly usage from October 19 to Sunday, up 801 percent from the previous week and ranking No.1 globally, according to OpenRouter data cited by business news site nbd.com.cn on Monday. OpenRouter is a global platform that aggregates access to LLMs.

Global usage of large language models reached 188 trillion tokens over the period, up 14.63 percent from the previous week, the data showed. 

Chinese AI models accounted for 93.54 trillion tokens, an increase of 18.14 percent week on week, while US models recorded 22.47 trillion tokens, up 29.44 percent. Chinese AI models have maintained a higher combined weekly token volume than their US counterparts for 26 consecutive weeks, according to the report.

Chinese AI models performed particularly strongly in the individual model rankings, occupying all of the top five positions in weekly token usage during the period. MiniMax M3.1 Flash (Free) ranked third with 13.7 trillion tokens. Zhipu AI's GLM-5.3 Flash came fourth with 13.2 trillion tokens, up 8 percent week on week, while Tencent's Hy4 Preview ranked fifth with 8.97 trillion tokens, an increase of 19 percent, according to data from OpenRouter.

The surge in usage of DeepSeek V4.1-Flash-1017 followed the company's announcement of a model update on October 17. Based on the V4.1 Flash model released on September 10, the updated version aims to improve performance in reasoning, coding, automation and vision-related tasks, according to the report.

Meanwhile, other Chinese AI companies are also accelerating model development and updates. MiniMax officially released its M3.1 Flash model on October 19. The company said the model offers improvements in complex agent tasks, computer operation, software development and cybersecurity. It claimed that the model's online inference speed had nearly tripled, while usage costs had fallen by 70 percent.

Chinese AI models have developed rapidly in recent years. The models have consistently driven innovation and created opportunities for global development, becoming integral digital infrastructure powering AI advancement worldwide.

For instance, Chinese companies have developed Arabic-language models to help Arabic speakers use digital technologies more conveniently through native speech generation. 

And, meteorological departments in more than 40 countries have used China's AI-driven MAZU model for early weather warning services via the cloud, with customized versions deployed in seven countries, including Pakistan, Ethiopia and the Solomon Islands. A long list of such cooperation endeavors proves that Chinese AI models are effective, safe and affordable, according to the Xinhua News Agency.