China's Daily Token Consumption Reaches 500 Trillion in June 2026
According to the China Academy of Information and Communications Technology (CAICT), China's daily average token consumption reached about 500 trillion in June 2026, marking nearly a fourfold increase compared to December 2025. This finding comes alongside the release of the 'Large Model Token Service Performance Monitoring Report (2026)' which evaluates public cloud LLM services. This rapid growth highlights the massive expansion of commercial AI adoption and Model-as-a-Service (MaaS) platforms in China. The performance benchmarks provide crucial data for enterprises struggling with model selection and service optimization. The report notes that domestic model throughput (TPS) doubled to an average of 78.23 tokens per second in August 2026, though it still trails behind typical foreign models averaging 112.82 TPS. Furthermore, context windows have expanded significantly, with 47.8% of monitored models supporting 1 million tokens.
## BACKGROUND
Model-as-a-Service (MaaS) provides artificial intelligence models through cloud APIs, allowing businesses to integrate advanced capabilities without managing underlying hardware. Metrics like Time to First Token (TTFT) and Tokens Per Second (TPS) are critical performance indicators used to measure inference speed and responsiveness in large language models.