On October 25, the China Academy of Information and Communications Technology, in collaboration with other institutions, issued the Large Model Token Service Performance Monitoring Report (2026). The report projected that by June 2026, the daily average token consumption in China is anticipated to surge to approximately 500 trillion, nearly quadrupling the figure recorded in December 2025. Drawing on data from January to August 2026, the report conducted a comparative analysis of model performance between original and third-party platforms, including DeepSeek, Qwen, GLM, and Kimi. The findings revealed that in August, the average TPS (tokens output per second) of domestic models hit 78.23 tokens/second, doubling from the 2025 average of 38.25 tokens/second. However, this still lagged behind the 112.82 tokens/second achieved by typical foreign models. Regarding service pricing, the normalized average price for domestic model services climbed to 13.46 yuan in 2026, nearly doubling from the previous year, yet remained markedly lower than the foreign average of 117.8 yuan.
