Tmall's Double 11 sale puts Chinese LLM token packages at 50% off or deeper

On November 11, Tmall's Double 11 festival featured an unusual category of discounted goods: token packages from large-model vendors. Reports from ifeng.com and 36Kr said packages from vendors including Zhipu, MiniMax and Qwen Office were marked down to 50 percent off or more.

The product on sale was not hardware or a membership perk but metered model usage. Packaging token-billed API capacity into an e-commerce bundle and placing it in a festival discount aisle is still a relatively new scene in the commercialisation of large models.

The vendor list points to several of China's main model providers, spanning general-purpose models and office-oriented product lines. The discount is described as 50 percent off or deeper, meaning some packages are being sold for less than half their list price.

The timing is easy to read. Double 11 is the year's most concentrated window of e-commerce traffic and a moment when both enterprise buyers and individual developers make purchasing decisions. Putting inference credits into that setting amounts to consumer-style promotion aimed at developers and small teams.

For developers, the immediate signal is that unit inference costs can fall sharply during a defined window; for vendors, the key question is whether the usage bought with those discounts turns into durable paying demand. A promotional price is not a long-term price, but it does reinforce expectations that model-inference pricing keeps drifting downward.

The reports so far cover only the discount depth and the participating vendors. They do not specify quota sizes, purchase limits or an end date for the promotion. Those details determine the real effective cost of the discount, and they are what would distinguish a short traffic-driven campaign from the opening move in a longer pricing strategy.

What to watch next: whether the discounts survive past Double 11, and whether vendors convert them into more permanent price cuts. If usage drops off once the promotion ends, this looks like a festival marketing push; if it sticks, it becomes another notch in the downward repricing of Chinese model APIs.