Skip to content
World clockEU--:--UK--:--USA--:--CN--:--PLDEFRIT中文EN

portal about AI and technologyevents · analysis · interviews · technical background

Search
LIVE
›

Doubao moves to subscriptions: 68, 200 or 500 yuan a month

ByteDance's assistant, the most popular AI app in China, has added three paid tiers alongside a free basic version. Behind the move sits a bill nobody wants to show in public: the cost of serving 120 trillion tokens a day.

Media & internetAnalysisRachel NwosuPublished: 21 September 20264 min readSources 2
Doubao moves to subscriptions: 68, 200 or 500 yuan a month

ByteDance says Doubao, its AI assistant and the most widely used app of its kind in China, is getting a professional version with three forecast tiers: 68, 200 and 500 yuan per month, roughly 37, 110 and 275 zloty. The basic version stays free, with no limit on usage time.

The news drew an immediate reaction. A Weibo thread about the paid Doubao subscription passed 330 million views. Users split into two camps. Some saw the fee as natural, since customers of ChatGPT or Kimi already pay for similar setups. Others said they would switch to a competitor.

Why free stopped paying off

The reason is arithmetic. The outlet TMTPost estimates that Doubao now handles more than 120 trillion tokens a day, up roughly a thousandfold since the product launched. At that scale, the industry puts the monthly cost of serving model inference somewhere between several hundred million and as much as a billion yuan. No revenue comes in, so it is a pure operating expense.

The financial pressure on the parent company is real. According to figures cited by Chinese media, ByteDance net profit in 2025 fell by more than 70 percent year on year, largely because of investment in AI infrastructure. The old model of "win users for free, worry later" ran out of road at the same time as investors ran out of patience.

Kimi already tested it

The pattern has already been tried on the Chinese market. The Kimi assistant, developed by Moonshot AI, earlier introduced tiers at 49, 99 and 199 yuan per month. Zhipu raised API access prices three times within a single year. The market is learning that generating answers for hundreds of millions of users cannot be funded by advertising. A search engine can carry ads. An assistant cannot, not without ruining the product.

The comparison with the West looks similar. ChatGPT passed a billion monthly active users, with conversion to the paid Plus plan at just 1.55 percent. That shows how thin the margin is between the free base and the paying tail.

Who will pay for it

Chinese commentators point to three possible revenue sources. The first is subscriptions for professional users: developers, students, content creators, office workers. The second is commerce built into the assistant. Doubao is to work with e-commerce partners, and the conversion rate in tests exceeds 3 percent, which for sellers is value in itself. The third is selling corporate access, including integration with editorial and marketing systems.

Two conclusions follow for media. First, access to AI assistants is ceasing to be a free distribution channel. If content is to be visible and cited inside them, someone will have to pay for it. Second, if the largest Chinese player openly calculates the cost of handling every query, pressure to commercialise publishers' content will not go away. Quite the opposite, it will grow.

Comments 0

Sources

2
  1. 01豆包推出专业版:68元、200元、500元三档订阅ZH
  2. 02QuestMobile 2026年6月AI应用月活排行ZH

All figures and quotations in this text come from the sources listed below.

Content prepared by the editorial team with AI assistance.

Rachel Nwosu

Rachel Nwosu

AI, models and technology

Rachel Nwosu covers AI, models and technology for FLASH24, working from public model documentation, benchmark releases and repository histories rather than press summaries, and she skips announcements that arrive without reproducible numbers. She checks training-data claims against dataset cards and reruns reported metrics where code is available. She spends much of her week interviewing researchers and engineers, tracking model launch calendars, and comparing vendor benchmarks with independent evaluations. Outside the desk she runs 3D printers, restores old computers, and tests how models learn from internet junk. She does not publish benchmark figures she cannot trace to a source.

Newsroom →

Comments

0
  1. No comments yet — be the first.

Write a comment

Comments are public. We do not publish abuse, spam or advertising.