Data from OpenRouter shows that global AI large model call volumes reached 56.8 trillion tokens last week (July 27 to August 2), a 2.07% decline from the previous period.

Among tracked models, Chinese AI large models accounted for 28.13 trillion tokens in weekly calls, down 14.76% week-on-week, while US AI large models recorded 4.38 trillion tokens, a surge of 87.18% from the prior week. Chinese large models have now led global weekly call volumes for 14 consecutive weeks.

Notably, all five of the top-ranked models by global call volume were Chinese products. DeepSeek-V4-Flash climbed to first place with 7.22 trillion tokens, up 13% weekly. Xiaomi MiMo-V2.5 ranked second with 6.3 trillion tokens, down 40%. Tencent Hy3 took third with 4.82 trillion tokens, up 22%. DeepSeek-V4-Pro rose to fourth with 3.28 trillion tokens, up 3%. Zhipu GLM-5.2 secured fifth with 2.89 trillion tokens, a 12% decline.

OpenAI's GPT-5.6 Luna entered the chart for the first time at eighth place, reaching 1.95 trillion tokens, a 456% weekly increase. OpenAI publicly released the GPT-5.6 Sol, Terra, and Luna versions on July 10, with Luna being the fastest and cheapest model in the series, ideal for high-frequency, large-scale tasks supporting tool calls and multi-step workflows. In less than three weeks, OpenAI announced pricing and performance adjustments to the GPT-5.6 series to improve cost-effectiveness for enterprise users, including an 80% price cut for GPT-5.6 Luna.

Analysts suggest OpenAI faces multiple competitive pressures. Chinese open-source models are rapidly iterating, narrowing the performance gap while widening the cost gap, forcing US frontier models to reconsider value. In mid-July, Moonshot AI released Kimi K3, matching or exceeding the results of closed-source US models on several benchmarks. Its model weights are fully open, allowing developers to deploy it on their own infrastructure and bypass API fees. Reports citing AP News note that Chinese models are now "almost as intelligent" as OpenAI's frontier models but at a fraction of the cost鈥?Kimi K3 charges just a few cents per million tokens compared to US counterparts.

Meanwhile, tech competitors like Google and Microsoft continue launching cost-efficient AI models, intensifying market price competition.

Disclaimer: This content and data are for reference only and do not constitute investment advice. Verify before use. Investing based on this information carries risks.