China's 2026 LLM Token Report: Prices and Throughput Double
A new report from the China Academy of Information and Communications Technology shows that domestic LLM daily token consumption surged to 500 trillion in June 2026, while average prices and output throughput both doubled compared to the previous period. This report provides crucial market insights into China's AI industry, highlighting how rising token demand and post-price-war pricing adjustments are shaping the Model-as-a-Service (MaaS) ecosystem. The report analyzes data from January to August 2026, showing domestic throughput reached 78.23 tokens per second (TPS) while average normalized pricing rose to 13.46 RMB, and models featuring a 1 million context window accounted for 47.8% of the monitored total.
## BACKGROUND
Model-as-a-Service (MaaS) is a cloud delivery model where pre-trained artificial intelligence models are provided to users through standardized APIs. Performance metrics like TPS (Tokens Per Second) measure generation throughput, while TTFT (Time to First Token) measures the initial latency before the first word is generated.