Chinese AI Models Overtake US in Weekly OpenRouter Usage for Second Straight Week
Chinese large language models extended their lead over US rivals on OpenRouter, with weekly usage climbing to 4.69 trillion tokens in the week from March 9 to March 15, up 11.83% from the prior week, according to the platform’s latest data. US models recorded 3.294 trillion tokens, down 9.33% week on week, marking the second consecutive week in which Chinese models surpassed US models in weekly calls.
The shift was most visible at the top of the global rankings, where the three most-used models were all Chinese.
MiniMax M2.5 held the No. 1 position for a fifth straight week, logging 1.75 trillion tokens despite a 6% decline from the prior week. StepFun’s Step 3.5 Flash (free) jumped into the top three for the first time at No. 2, with weekly usage of 1.34 trillion tokens, up 79% week on week. DeepSeek V3.2 ranked third with 1.04 trillion tokens, a 25% increase.
The data also highlighted the emergence of a previously little-known model. Hunter Alpha debuted in the rankings at No. 7 with 0.666 trillion tokens. OpenRouter’s model description says Hunter Alpha went live on March 11 and is designed for agentic applications, featuring more than a trillion parameters and a roughly 1 million-token context window. The listing says it is aimed at long-horizon planning, complex reasoning and sustained multi-step task execution, with reliability and instruction-following accuracy intended to meet requirements for frameworks including OpenClaw.
For investors tracking the pace of commercial adoption in generative AI, OpenRouter’s usage figures point to where developer demand and application traffic are concentrating week to week. The sharp rise for Step 3.5 Flash (free) alongside steady high volumes for established leaders suggests that usage leadership is being influenced not only by entrenched incumbents but also by rapid re-ranking as new releases gain traction.
Related Coverage: