China's Daily AI Token Consumption Hits 500 Trillion, Nearly 4x in Six Months

The China Academy of Information and Communications Technology and partner organizations have released a 2026 report on large model token service performance, based on monitoring data from January to August.

The report shows that in June 2026, China's daily token consumption reached about 500 trillion, nearly 4 times the December 2025 level. In August, Chinese models averaged 78.23 tokens per second, double the 2025 figure, while leading international models averaged 112.82. The time to first token rose to 1.26 seconds, linked to bigger models, longer context and higher usage.

Prices have also climbed. The normalized average price of Chinese models rose to 13.46 yuan, nearly doubled, while typical international models cost about 117.8 yuan. Models with 1 million-token context now account for 47.8%. The open-source models DeepSeek, Qwen, GLM and Kimi show clear performance gaps across platforms, with output speed differing by up to nearly 2x, mainly due to different optimization strategies.