It was neither GPT nor Claude. The AI (artificial intelligence) model most used by developers worldwide turned out to be Xiaomi's 'MiMo-V2.5'. Meanwhile, in AI performance evaluations, the latest models from OpenAI and Anthropic maintained their top positions. As Chinese AI companies rapidly increase actual usage by leveraging price competitiveness, a clear competitive structure is emerging where U.S. companies compete by highlighting superior performance.
According to the latest AI model rankings released on the 29th by the AI model platform OpenRouter, Xiaomi's MiMo-V2.5 ranked first in token usage over the past week. MiMo-V2.5 processed 10.8 trillion tokens during the week, accounting for approximately 15% of the total.
Second place went to DeepSeek's DeepSeek V4 Flash (6.73 trillion tokens), and third to Tencent's Hunyuan3 (4.76 trillion tokens). Following them were DeepSeek's DeepSeek V4 Pro (3.34 trillion tokens) and Z.ai's GLM 5.2 (3.31 trillion tokens). The top five models were all held by Chinese companies, indicating their growing presence in the developer market.
However, this ranking is based on the actual API (application programming interface) token usage processed through the OpenRouter platform. While it does not reflect the total number of users for ChatGPT or Claude, it holds significance as an indicator showing which models developers building AI services are actually choosing.
The rapid rise of Chinese AI models is attributed to their price competitiveness. Industry analysis suggests that Chinese AI companies are quickly securing developers by offering relatively lower API fees and higher cost-performance ratios compared to the latest U.S. models. For companies operating AI services, operational costs are as important as performance; thus, there is a growing trend of adopting Chinese models that provide performance above a certain threshold while keeping cost burdens low.
In contrast, U.S. companies maintained their lead in AI performance. In OpenRouter's comprehensive benchmark, Anthropic's Claude Opus 5 recorded the highest score of 60.7 points. It was followed by Claude Fable 5 (59.9 points), OpenAI GPT-5.6 Sol(max) (58.9 points), Moonshot AI Kimi K3 (57.1 points), and Claude Opus 4.8 (55.7 points). OpenAI's GPT-5.5(xhigh) also secured seventh place with a score of 54.8 points. While Chinese models dominate in usage, U.S. models lead in performance, signaling the full-scale launch of strategic competition between 'cost-performance' and 'top-tier performance' in the AI market.
in TECH
