The 2 Trillion Token Illusion

“3-2-1, Token goes live.”

On the evening of November 2, 2026, that line ran through Li Jiaqi’s livestream comments. The product link on screen cost 0.01 yuan. Half an hour later, 100,000 Qwen Office personal monthly cards were gone, roughly 2 trillion tokens in quota. Chen Yusen, Alibaba Group vice president and CEO of Qwen Office, sat beside Li Jiaqi. He demoed AI organizing homework, building a training game, and analyzing livestream replays. The comments moved fast. The numbers looked huge.

The sales figure did not. 100,000 times 0.01 yuan is 1,000 yuan.

The livestream had little to do with revenue. It was a channel experiment built around renewal rates, dressed up as a trillion-token headline.

20 million tokens is a ceiling, not a burn rate

Each card carries about 20 million tokens: 500 base credits, 500 monthly bonus credits, and 300 new-user credits, converted under Qwen Office’s redemption rules. Multiply by 100,000 and you get 2 trillion.

That 20 million is a theoretical quota. It is not the compute a user will necessarily consume. Qwen Office runs on credits. The exchange rate between credits and different models or task types is not uniform. Opening a page, reading an attachment, generating an image, and sorting plain text each have their own consumption logic. Turning credits into tokens and then into “trillions” works as a headline. It does not work as business analysis.

Tokens are also not a standard unit across models. Tibo, who leads OpenAI Codex, has noted that the same passage can split into anywhere from 766 to 1,170 tokens depending on the tokenizer, a gap above 50%. After Claude’s later models switched tokenizers, the same input produced about 30% more tokens than earlier models. Even if a user burns all 20 million tokens, the work completed depends on the model called and the task performed. A livestream impulse buy cannot tell a consumer what quality of output those tokens will buy.

A product with no standard price tag is a poor candidate for stockpiling.

An enterprise product wearing a consumer costume

Qwen Office was not built for a livestream. In July 2026, Alibaba merged QoderWork, Wukong, and MuleRun into Qwen Office and made it a first-level business unit under ATH. The target was enterprise AI productivity: organizations, workflows, seat-based billing. The enterprise standard edition costs 198 yuan per seat per month and supports VPC deployment and unified management. First-month users passed 30 million, and enterprise users made up more than half. That is a B2B path.

In Li Jiaqi’s studio, Chen Yusen talked about keeping cats and dogs company, caring for family, and leaving work on time. The product on sale was the personal basic monthly card, list price 28 yuan. The demos were homework and a fishing map. The enterprise story disappeared. In its place was consumer language: saves time, fun, low barrier.

The consumer scenario has value. The problem is fit. The users activated by this livestream and the enterprise customers Qwen Office needs for a commercial loop are not in the same coordinate system.

The pricing makes the gap wider. The 0.01 yuan card is the basic edition. Qwen Office’s personal paid tiers run from 78 to 158 yuan per month: 78 yuan for standard continuous billing, 158 yuan for advanced. Between 0.01 and 78 lies a fault line in usage habits and willingness to pay. Of the 100,000 users acquired almost free, how many will stay at the full price in month two? That is the experiment’s reading.

The channel wanted a translator and a funnel

Direct sales revenue makes the event look absurd. Put it inside Alibaba’s AI office competition and its Token e-commerce buildout, and the logic appears.

First, user education. AI office products usually reach app stores, developer communities, and enterprise sales teams. Those audiences already know the category. Li Jiaqi’s livestream can put the word “Token” in front of hundreds of thousands of non-technical users in half an hour. Zhang Yi of iiMedia Research called it a C-end market enlightenment movement under the Token economy. The intent matters more than the token count. Pan Helin of the Ministry of Industry and Information Technology offered the other side: most Token demand still sits with developers and enterprises, while livestream audiences are consumers, so this leans toward a test. Both readings hold. The enlightenment is real. The test is shallow.

Second, Alibaba is building a Token retail channel. On September 3, 2026, Tmall launched a Token recharge center called AI Space Station. It connected Alibaba Cloud, Zhipu, Kimi, and MiniMax, with subscription packages and pay-as-you-go options. The 0.01 yuan card is the loudest anchor on that channel line. The same livestream also listed AI computers preloaded with Qwen Office, AI recording cards, and desktop robots. Token works like ink with a printer. The hardware entry point and the subscription habit are the sale. Token is the hook that lowers the trial barrier.

A channel experiment succeeds or fails at the bottom of the funnel, not at the entrance. Selling a card, redeeming it, starting the first task, and renewing after a month are four different actions. 0.01 yuan buys the order. It does not buy redemption, usage, or renewal. The livestream proved that AI products can use traditional e-commerce to reach users. It did not prove that reach becomes retention.

Token will not become a hot product. The token hook will keep coming back.

Will AI tokens become a livestream bestseller?

As a standalone category, probably not. Livestream commerce runs on low price, impulse, and instant gratification. Token value depends on model capability, context length, tool-calling stability, and agent orchestration. A 0.01 yuan price tag cannot carry that information. A product whose value appears after a month of use does not fit “3-2-1, link goes live.” Expecting people to stockpile tokens like lipstick is like expecting them to stockpile air. At least air has a standard unit.

As a channel strategy, the token hook will return. The likely forms are hardware plus token bundles, service plus token packages, and a cheap first month followed by a normal-price renewal. Token will not be hoarded like lipstick. It will work like printer ink: a bundled entry point and an ongoing revenue stream for hardware and services. The product that mattered in this livestream was never the 2 trillion tokens. It was 100,000 accounts pulled in by 0.01 yuan, waiting to see whether they convert.

Qwen Office’s long-term value sits on the enterprise side, not in a livestream. But enterprise AI competition is moving from parameters and leaderboards toward everyday user habits. Letting an ordinary consumer hear “Token” for the first time and try an AI office tool for 0.01 yuan has a certain logic. The open question is what happens after the introduction. Can the product hold the people attracted by the low price? Will they pay 78 yuan in month two or three for an AI tool with no Li Jiaqi shouting “link goes live”?

That is the yardstick for Token e-commerce. Not a night measured in trillions.