China Academy of Information and Communications Technology (CAICT) said in its
2026 Large Model Token Service Performance Monitoring Report that domestic daily
token consumption was about 500 trillion in June 2026, nearly four times
December 2025. Using Jan–Aug 2026 monitoring data, the report compared
original-vendor and third‑party platforms (including DeepSeek, Qwen, GLM, Kimi).
In August the average across domestic models was 78.23 tokens/sec (TPS), roughly
double the 2025 average of 38.25 TPS, but below a representative foreign model
at 112.82 TPS. Normalized domestic model service prices have risen since early
2026 to 13.46 yuan, up almost twofold, versus a foreign average of 117.8 yuan.
