Runjian Co.'s subsidiary Wuxiang Yungu signed a US$200 million cross-border AI token sales contract with an overseas company X, with computing power deployed domestically, overseas API calls, and settlement based on token consumption.
A CAICT report shows that China's normalized average token price rose from RMB 6.8 per thousand tokens in the second half of 2025 to RMB 13.46 so far in 2026, while average daily token consumption grew about threefold over the same period.
The AGI-Enabled Token Economy Application Innovation Alliance was unveiled in Chongqing, and the Mobile Cloud (Chongqing) Token Center was launched at the same time; Chongqing has included the token economy in its 15th Five-Year Plan for AI development and is exploring pay-per-use token billing.
MiniMax open-sourced M3.1 Flash, which processed over 100 trillion tokens within 72 hours of launch; API pricing is RMB 1 per million input tokens and RMB 4 per million output tokens, more than 50% lower than the previous M3 for both input and output.
The Chengdu Token Innovation Consortium was unveiled, iSoftStone's "Beijing **** Token Factory" went into production, and the Greater Bay Area Token Factory came online, pushing computing services from cabinet leasing toward token-based supply.
Wang Jiandong of the State Information Center proposed that Token business models will move from pay-per-use and monthly subscriptions to "value-based pricing," while noting that tokens are not comparable across different models and that Token loans and Token futures mostly remain concepts.
Ramp data shows that about 80% of OpenAI's and Anthropic's enterprise revenue comes from top customers accounting for only 1% of total customers; enterprise token spending was almost flat in August and September, and in early October it was only 1.5% above the August peak.
Anthropic's actual token price index fell from above US$1.70 in mid-September to US$0.83, a drop of more than 50%; OpenAI's corresponding index fell to US$0.45, deepening price deflation among leading models.
Claude Opus 5.5 is priced at US$4 per million input tokens and US$20 per million output tokens, while GPT-6 Sol is priced at US$2 and US$10; one independent index shows Opus consuming 260M output tokens per task at a cost of US$5.98, versus Sol's 77M at US$1.04.
Moonshot AI's Kimi K3.1 Pro entered the competitive band with a 1 million-token context window and pricing of US$2 per million input tokens and US$10 per million output tokens, about half of Claude Opus 5.5.