According to OpenRouter data published on September 7, 2026, Tencent's Hunyuan Hy4 Preview topped the global large language model ranking for the week of August 31 to September 6, with weekly token usage of 14.7 trillion tokens, up 379 percent week-on-week.
Chinese AI large models recorded total weekly usage of 56.72 trillion tokens, up 2.83 percent week-on-week. By comparison, US AI large models totaled 16.54 trillion tokens, down 3.1 percent week-on-week. This marked the 19th consecutive week that Chinese model usage exceeded US usage globally.
Four of the top five global models in the ranking were Chinese. Zhipu GLM-5.3 Flash rose to third place with 12.4 trillion tokens, up 101 percent. DeepSeek-V4-Flash-0731 (the official V4 Flash release) ranked fourth with 12.4 trillion tokens, and DeepSeek-V4-Flash-0423 (the preview version) ranked fifth with 5.19 trillion tokens.
The Hy4 Preview was released and open-sourced on August 28, with 770 billion total parameters, 49 billion activation parameters, and a context length of more than 1 million tokens. On September 1, Tencent released a lightweight version that compresses model weights from 1.5 TB to 214 GB, dramatically lowering deployment thresholds.
Source: National Business Daily / Sina News
