The latest weekly AI large model usage statistics from the global multi-model aggregation platform OpenRouter show that the top five products are all developed by Chinese companies.
Xiaomi's MiMo-V2.5 secured the top spot with a weekly usage volume of 10.5 trillion tokens, marking a 12% increase from the previous week. DeepSeek placed two models in second and fifth positions, Tencent's Hunyuan Hy3 ranked third, and Zhipu AI's GLM-5.2 came in fourth. The data indicates that Chinese AI large models are maintaining a strong lead in global usage volume.
According to public information, MiMo-V2.5, which topped the list, was officially launched for public testing on April 23 and its full series was open-sourced by the end of that month. Xiaomi reported that the model's weekly usage volume grew from 1.5 trillion tokens to 10.5 trillion tokens, increasing more than sixfold in two months, and it achieved the top spot in both weekly and monthly rankings. Key users include independent developers, startup teams, and small to medium-sized enterprises involved in AI programming and agent development. These users prioritize model inference cost, deployment efficiency, and stability, leading to rapid growth for open-source models with lower usage costs and strong agent capabilities.
A look at the listed products reveals that multiple Chinese large models have adopted differentiated strategies. DeepSeek's flagship version is designed for code generation and complex agent tasks, with performance on par with mainstream international closed-source models. Tencent's Hunyuan Hy3, which was open-sourced in early July, saw a sharp increase in usage, with a week-over-week growth rate of over 999%. Zhipu AI's GLM-5.2 also maintains high activity levels among developers. Several companies are embracing open-source strategies, which lowers the barrier for developers and promotes the model's application adoption.
Beyond usage volume, the ecosystem for Chinese open-source models is also expanding. Public data shows that on the global open-source AI community Hugging Face, Chinese open-source models rank among the top globally in cumulative downloads. The Qwen series leads the list with over 1 billion downloads. Li Wei, Deputy Director of the Cloud Computing and Digital Research Institute at the China Academy of Information and Communications Technology, noted that domestic models use open-source strategies to activate the innovation ecosystem, leveraging superior performance, controllable costs, and ease of local deployment to significantly lower the barrier to AI technology application.
According to data from U.S. investment institutions, approximately 80% of American AI startups use Chinese open-source models during their funding roadshows, with over 200 derivative models based on Chinese large models emerging every day. Industry experts believe that as China continues to advance large model research and development, open-source ecosystem construction, and industrial applications, Chinese large models are transitioning from single-point technological breakthroughs to a new phase of coordinated development across technology, industry, and the ecosystem. Currently, large models are being progressively applied in areas such as AI programming, agent development, and office collaboration, with the developer ecosystem and application scenarios continuously expanding. Leading in traffic is just the starting point; the domestic AI industry still needs to make persistent efforts in multiple dimensions, including independent computing power, the underlying ecosystem, and cutting-edge innovation, to consolidate market advantages through long-term technological accumulation.