Home
China’s large language model usage now exceeds that of global competitors, with MiniMax M2.5 leading the field

Based on the most recent tracking information provided by OpenRouter, the artificial intelligence industry in China is showing rapid expansion. From March 2 through March 8, the total weekly usage volume of large-scale models across China reached 4.19 trillion Tokens, marking a substantial rise of 34.9% compared to the prior week. This figure has now exceeded that of the United States, which recorded 3.63 trillion Tokens, for two consecutive weeks.
Within the global ranking of the top five countries by large model usage volume, Chinese companies have performed exceptionally well, securing three spots in the list.
Key details from this ranking:
MiniMax leads globally: MiniMax’s M2.5 model retained its position at the top of the global rankings, with a weekly usage volume of 1.87 trillion Tokens—up 15% from the previous week—highlighting its strong market dominance.
DeepSeek shows stable performance: DeepSeek V3.2 finished in third place globally, accounting for 0.83 trillion Tokens in weekly usage volume.
Stepwise Star rises unexpectedly: Step3.5 Flash emerged as a major surprise this week, experiencing a sharp increase of 69% in weekly usage volume to reach 0.75 trillion Tokens, thereby breaking into the global top five.
Kimi slightly declined: Kimi K2.5 dropped to sixth place, with a weekly usage volume of 0.74 trillion Tokens—up 8% from the previous week.
Industry experts suggest that China’s continued leadership in large model usage volume is driven by high activity within the domestic AI application ecosystem and ongoing improvements in computing costs. As more companies integrate their operations with locally developed large models, China’s strengths in AI development and application continue to grow stronger.
Related article
Google Unveils Googlebook: Gemini Model Redesigns Pointer Interaction and System Infrastructure
Google has officially introduced Googlebook, a new AI notebook computer platform that redefines personal computing by shifting from traditional operating systems to a model-centric architecture. By deeply integrating the Gemini large language model a
How to fix Core Web Vitals for better Google ranking?
Local vs. National SEO for Lawyers: Key DifferencesThe Internet has become the primary resource for individuals seeking legal assistance. Whether it's a quick question about tenant rights or the need for a high-powered corporate attorney, most search
AntGroup Unveils Ling-3.0-flash-VL, a 124B-Parameter Native Multimodal Large Model
The Ant Open Source team has officially launched Ling-3.0-flash-VL, the first native multimodal large model in the Bailing series. Built on the MoE architecture of Ling-3.0-flash, this model features 124B total parameters, with only 5.5B activated pe
Related Special Topic Recommendations
Comments (0)
0/500

Based on the most recent tracking information provided by OpenRouter, the artificial intelligence industry in China is showing rapid expansion. From March 2 through March 8, the total weekly usage volume of large-scale models across China reached 4.19 trillion Tokens, marking a substantial rise of 34.9% compared to the prior week. This figure has now exceeded that of the United States, which recorded 3.63 trillion Tokens, for two consecutive weeks.
Within the global ranking of the top five countries by large model usage volume, Chinese companies have performed exceptionally well, securing three spots in the list.
Key details from this ranking:
MiniMax leads globally: MiniMax’s M2.5 model retained its position at the top of the global rankings, with a weekly usage volume of 1.87 trillion Tokens—up 15% from the previous week—highlighting its strong market dominance.
DeepSeek shows stable performance: DeepSeek V3.2 finished in third place globally, accounting for 0.83 trillion Tokens in weekly usage volume.
Stepwise Star rises unexpectedly: Step3.5 Flash emerged as a major surprise this week, experiencing a sharp increase of 69% in weekly usage volume to reach 0.75 trillion Tokens, thereby breaking into the global top five.
Kimi slightly declined: Kimi K2.5 dropped to sixth place, with a weekly usage volume of 0.74 trillion Tokens—up 8% from the previous week.
Industry experts suggest that China’s continued leadership in large model usage volume is driven by high activity within the domestic AI application ecosystem and ongoing improvements in computing costs. As more companies integrate their operations with locally developed large models, China’s strengths in AI development and application continue to grow stronger.
Google Unveils Googlebook: Gemini Model Redesigns Pointer Interaction and System Infrastructure
Google has officially introduced Googlebook, a new AI notebook computer platform that redefines personal computing by shifting from traditional operating systems to a model-centric architecture. By deeply integrating the Gemini large language model a
How to fix Core Web Vitals for better Google ranking?
Local vs. National SEO for Lawyers: Key DifferencesThe Internet has become the primary resource for individuals seeking legal assistance. Whether it's a quick question about tenant rights or the need for a high-powered corporate attorney, most search
AntGroup Unveils Ling-3.0-flash-VL, a 124B-Parameter Native Multimodal Large Model
The Ant Open Source team has officially launched Ling-3.0-flash-VL, the first native multimodal large model in the Bailing series. Built on the MoE architecture of Ling-3.0-flash, this model features 124B total parameters, with only 5.5B activated pe











