June 8, 2026 – Global AI model usage hit a fresh milestone last week, with total token consumption soaring to 36.1 trillion between June 1 and 7 — a 13.5% jump from the prior week, marking seven consecutive weeks of growth, according to new data released by OpenRouter.
China’s AI sector continues to outpace the rest of the world. Chinese models processed a staggering 14.19 trillion tokens during the same period, up 27.49% week-over-week and rising for the third straight week. That figure puts China firmly ahead of the United States, which saw usage slide 24.53% to just 3.2 trillion tokens. China has now topped the U.S. in weekly AI call volume for six consecutive weeks.
The shift is even more pronounced at the model level. Of the top five most-used models globally last week, four were Chinese-developed.
DeepSeek-V4-Flash held its crown for the third week running, racking up 3.69 trillion tokens — a 19% increase from the previous week. Tencent’s Hy3 preview held steady at number two with 2.94 trillion tokens, though it dipped 3% week-over-week.
A fresh face made an impressive debut: MiniMax’s newly launched M3 model burst into the top three with 2.5 trillion tokens in its very first week. It is the first domestic model to combine cutting-edge coding performance, a 1-million-token context window, and native multimodal capabilities in a single package.
Xiaomi’s MiMo-V2.5 claimed the fourth spot, generating 2.19 trillion tokens — a sharp 50% surge from the week before.
Meanwhile, Anthropic’s Claude lineup — long a staple in the global developer ecosystem — showed clear signs of losing ground. Both Claude Sonnet 4.6 and Claude Opus 4.7 tumbled out of the top five, with usage dropping 9% and 38% respectively. For Sonnet 4.6, it marked the first time in two months it failed to crack the top five.
