Zhipu AI recently brought tokens to Tmall. In its Zhipu flagship store, the GLM Coding Plan has been packaged into several standardized products: personal Lite, Pro, and Max versions, plus a team standard seat, priced at RMB 118, RMB 538, RMB 1,078, and RMB 598 per month, respectively. The plans are based on GLM-5.3 and support more than 20 mainstream agents, including ZCode, Claude Code, and Codex. Zhipu AI also says that in the future users will be able to purchase tokens directly on e-commerce platforms.
A question follows: Zhipu AI already has its own MaaS platform, with more than 7.4 million enterprise and developer users as of the end of August, so why does it still need Tmall? The answer may lie in a change in how AI capabilities are consumed. For heavy developers, Tmall may be optional, but for light AI users and small-to-medium developers, Tmall packages complex technical terminology into products that are understandable and purchasable, allowing on-demand buying much like topping up phone credit.
More noteworthy is the change in token consumption structure driven by agents. A coding agent will continuously read code, modify files, run tests, and analyze errors. The model no longer answers a question just once; it is invoked continuously across a complete task, creating high-frequency, recurring consumption. This is the underlying logic behind Zhipu AI’s bet on AI Coding over the past year: turning one-off Q&A into ongoing consumption.
The financial numbers confirm the shift. In the first half of 2026, Zhipu AI’s revenue reached RMB 954 million, up 399.7% year-on-year. Open platform and API business revenue was RMB 825 million, up 2,735.7% year-on-year, accounting for 86.5% of total revenue and replacing localized deployment as the largest revenue source. In the same period last year, that share was only 15.2%. As of the end of August, token calls on the MaaS platform had grown more than 40x since the beginning of the year, while paying daily active users increased 603%.
But hold the applause. Behind Zhipu AI’s rising API volume and price, concerns remain. As of the end of August, the average API selling price had increased by about 101% since the start of the year, clearly boosting revenue, but overall gross margin fell from 50% in the year-earlier period to 26.4% in the first half, with an adjusted net loss of RMB 1.964 billion. Token calls grew 40x, and expanding usage actually offset unit cost declines: the more users ask and the more complex the agents run, the greater the computing power spending on inference infrastructure.
Zhipu AI’s own solution is to lower the cost per call. Through domestic computing power, model optimization, and