Statistics from OpenRouter, a global AI model aggregation platform, show total global large-model calls reached 164 trillion tokens in the week of October 12-18 - and Chinese models' weekly call volume has now exceeded that of US models for 25 consecutive weeks, remaining firmly in first place globally.
The same week, the global top five models by call volume were all Chinese AI models, extending a streak that has turned usage share into one of the clearest indicators of where AI capability is being deployed at scale.
The data point lands amid a run of Chinese model news: Moonshot's 3-trillion-parameter open-source Kimi K3.1 Pro on October 24, iFlytek's domestically trained Spark X2.6 on October 22, and a widening open-source push that pairs frontier performance with prices a fraction of closed Western rivals.
Analysts read the trend as evidence that China's AI industry advantage is compounding not through any single benchmark win but through breadth - model diversity, open weights, low-cost inference and fast scenario deployment across industries.
Image: Efficiency of AI-related computer chips. Source: Wikimedia Commons (CC BY-SA 4.0).
