The weekly calls of China's large models have surpassed the United States for six consecutive weeks, and the top four are all domestic models.

According to the latest monitoring data from OpenRouter, the total number of global AI large model calls reached 36.1 trillion Tokens last week (June 1 to 7), a month-on-month increase of 13.5%, rising for seven consecutive weeks. The most eye-catching signal is that the weekly call volume of China's large models reached 14.19 trillion Token, a month-on-month increase of 27.49%, while the United States only had 3.2 trillion Token during the same period, a month-on-month decrease of 24.53%. China's weekly calls for large models have surpassed the United States for six consecutive weeks, ranking first in the world.

Top five: Domestic models dominate the list

Among the top five models globally called last week, the top four are all large Chinese models:

1. DeepSeek-V4-Flash — 3.69 trillion Tokens, a month-on-month increase of 19%, topping the list for three consecutive weeks
2. Tencent Hy3 preview — 2.94 trillion Tokens, ranked second for three consecutive weeks
3. MiniMax M3 — broke into the top three in the first week, with 2.5 trillion tokens, the first domestic model with cutting-edge programming capabilities, 1M ultra-long context and native multi-modality
4. Xiaomi MiMo-V2.5 — 2.19 trillion Tokens, a month-on-month surge of 50%
5. The fifth one is the American model

From a manufacturer perspective, the total number of DeepSeek models used reached 6.75 trillion Tokens, exceeding Anthropic and Google combined, ranking first among manufacturers for four consecutive weeks. The total number of MiniMax calls is 3.05 trillion Tokens, surpassing Xiaomi and Tencent.

Claude Series Slipping

It is worth noting that the Claude series, which has long occupied the core of the global developer ecosystem, has seen a significant decline. Claude Sonnet 4.6 and Claude Opus 4.7 both fell out of the top five, with weekly calls falling by 9% and 38% respectively. This set of data is in sharp contrast to the continued climb of the DeepSeek V4 series, indicating that open source/cost-effective models are systematically eroding the market share of closed source flagship models.

From the perspective of industry trends, the collective rise of Chinese models is not only reflected in the absolute leadership in the number of calls, but also in the formation of ecological diversity - DeepSeek's ultimate cost-effectiveness, Tencent's ecological integration, MiniMax M3's entry into the top three in its first week (the first domestic model with cutting-edge programming capabilities, 1M ultra-long context and native multi-modality), Xiaomi's aggressive pricing (MiMo-V2.5 surged by 50% month-on-month) - each of the four paths has its own emphasis. However, it is worth noting that Xiaomi MiMo-V2.5-Pro ​​achieved a 321% month-on-month increase in weekly call volume after a significant price reduction in the previous week. It has fallen off the list this week, showing that the pure price strategy has obvious fluctuations in the driving effect of call volume. Claude Sonnet 4.6 and Opus 4.7 both fell out of the top five (down 9% and 38% month-on-month respectively), sending a clear signal: In an era when developers vote with their feet, model capabilities are no longer the only determining factor, and price and accessibility are becoming equally important competitive dimensions.