The global landscape of artificial intelligence usage underwent a dramatic shift in late April, as Chinese large language models (LLMs) surged past their American counterparts in total weekly token volume. According to the latest data from OpenRouter, global AI call volume reached 23.9 trillion tokens during the week ending May 3, 2026, marking a nearly 9% increase. Within this total, Chinese models accounted for 7.942 trillion tokens—a massive 81.7% jump week-over-week—while American model usage dipped significantly to 3.258 trillion tokens.
Tencent’s Hy3 preview (free) led the charge, claiming the top spot with 3.03 trillion tokens in a single week. This represents a staggering 799% increase in usage, suggesting that the tech giant's strategy of offering high-performance models for free is successfully capturing a massive developer and user base. Following closely was Moonshot AI’s Kimi K2.6, which secured second place globally with 1.28 trillion tokens, further cementing China’s current numerical lead in the AI sector.
DeepSeek has emerged as a particularly disruptive force, with its V4-Flash model debuting in the top ten after a 344% increase in call volume. This rapid adoption is directly linked to an aggressive price war initiated by the firm. DeepSeek recently slashed prices twice in quick succession, bringing the cost of its V4-Flash model down to just 0.02 yuan per million tokens. This move has forced competitors to recalibrate their monetization strategies as intelligence becomes increasingly commoditized.
The divergence between Chinese and American usage patterns reveals a broader strategic shift. While American firms like OpenAI and Anthropic have focused on high-margin, sophisticated reasoning models, Chinese players are aggressively scaling through low-cost or free access. This high-volume approach serves as a critical data engine, allowing Chinese developers to iterate their models based on massive real-world feedback loops across diverse industries.