The report covers monitoring conducted from January through August 2026.
It reports that China’s normalized average token-service price rose from 6.8 yuan in the second half of 2025 to 13.46 yuan in 2026. Typical overseas models released in 2026 averaged 117.8 yuan on the same normalized basis. Daily domestic token consumption reached about 500 trillion tokens in June, nearly four times the December 2025 level.
For August, domestic models averaged 78.23 output tokens per second, versus a 38.25 average in 2025, while average time to first token rose from 0.78 to 1.26 seconds. The source says 47.8% of monitored models had one-million-token context windows. It also reports nearly twofold throughput differences for the same DeepSeek, Qwen, GLM, or Kimi model across third-party platforms.
The source does not establish the causes of price or latency changes, publish enough methodology here to independently reproduce the results, or show that one provider will outperform another for a particular workload.