AI companion app Zeta's monthly revenue tops $10 million, charging points per turn on top of membership

South Korea's Scatter Lab, through its Zeta app, has turned this into an explicit paid system: a monthly fee buys perks such as ad removal, while the advanced models consume points for each round of replies. Having the model revise an already-generated reply on command also carries its own separate point price.

In AppMagic's estimate for August 2026, Zeta's monthly revenue after app store commissions exceeded $10 million. What makes this business worth taking apart is how the platform splits a single companionship experience into different paid items, and how it keeps delivering the services users have paid for.

Zeta's basic chat is free. Users can subscribe for the usage environment, and can also buy points for generation and revision with the advanced models. By the end of October, the platform also put subscriptions on sale on the web and added automatic point top-ups. Around the same story, the models, creation tools and payment methods are being adjusted together.

Illustration of a multi-character interactive story

Membership fees do not cover every generation

Zeta Pass's monthly benefits include ad removal, faster character replies, unlimited creation of private content, and 200 Pieces. Pieces are in-app points, consumed by operations such as advanced model conversations and image generation. Membership benefits improve the usage environment; which model to choose and how many times to use it are a separate expense.

The two payments differ in actual price. In the Japanese App Store, Zeta Pass costs 1,650 yen per month, while 200 Pieces bought separately cost 300 yen and 1,000 cost 1,500 yen. The 200 points included with membership are worth only the smallest point pack; the 1,650 yen also buys a full month of membership benefits, so it cannot all be counted as the unit price of 200 points.

Starting February 25, 2026, Zeta changed the Pro service, previously purchased manually with points, into a separate monthly-renewing Pass. Points are no longer used to buy this monthly subscription; subscription accounts instead receive 200 points each month, with the bonus points valid until the next billing date. Membership renewals and model usage thus form two separate billing rhythms.

Zeta membership and points payment relationship

For the platform, this split means light and heavy users do not have to pay exactly the same service fee. One person chats only a few rounds a day, another pushes a very long plot forward continuously; both can buy the same membership, and the latter's heavier advanced-model usage continues to be charged through points. As usage grows, revenue has an entry point that grows with it.

The 200 points also connect the two charging systems. Members receive a small monthly allowance of advanced features, enough to start using them without buying extra points; once used up, whether to return to the free model or keep topping up becomes a more concrete choice. The platform separates the first payment from subsequent spending, and also places membership renewal and model spending separately before the user.

The longer the story, the more the platform must calculate service costs

Once a traditional novel is written, the same text can be read repeatedly. Stories in Zeta must keep generating with each user input: when the user changes a choice, the model must reorganize character reactions, prior context and the next passage. Settings written by creators can be used by different users, but the story each user sees next still has to run separately.

If only a fixed monthly fee were charged while advanced models were supplied without limit, the extra calls from heavy users would not bring corresponding per-use revenue. Charging points per round lets the platform price more expensive services separately and arrange different budgets for different intensities of use. The charging structure begins to correspond to the ongoing generation work, not just to a month's access entitlement.

On August 7, Zeta raised Koji's per-round price from 4 points to 5 and Luca's from 8 to 10, both up 25%. The adjustment involved the end of a Luca promotion and an increase in the amount of information the two models reference when generating replies. As the platform expanded the information the models must process, it also adjusted the usage price.

For users, the change is more direct: if all 200 points are used for conversation, the number of rounds Koji can generate falls from 50 to 40, and Luca from 25 to 20. For the same allowance, one-fifth fewer replies can be bought. At the Japanese App Store's 200-point, 300-yen tier, one Koji generation works out to 7.5 yen and one Luca generation to 15 yen.

A fixed per-round price is easy to understand, but it also packs tasks of different lengths into the same billing unit. Simple small talk and a plot that must manage relationships among many characters are both charged per round under the same model. The platform still has to arrange how much prior context to read each round, which settings to invoke, and how long a reply to generate; context management and model efficiency affect the computing resources behind that fixed price.

So raising revenue cannot rely only on attaching higher prices to stronger models. If less prior context is read to control consumption, character relationships and plot continuity may suffer; if processing volume is increased to maintain the experience, the cost of the service changes again. Per-use charging connects revenue to usage volume, and the product team must also make each round of generation both deliverable and sustainably supplied.

Charging for revisions puts generation quality on the bill

Zeta's "magic wand" lets users issue instructions for the model to revise a reply. The point rules after August 7 are: one revision with the free zeta model costs 5 points, Koji 5 points, and Luca 10 points. For the latter two models, revising a reply and generating a new round of replies cost the same number of points.

Take Luca as an example: generating 10 points at a time, then immediately revising once costs another 10 points. If all 200 points are used only for generation, you get 20 rounds of replies; if each round is generated and then revised, you can complete only 10 sets of operations. Converted to Japanese in-app point prices, one set of operations is 30 yen. At this point, what users are weighing is no longer just the number of model calls, but how much the story actually advances after spending that money.

Premium model points and operation quotas

Revisions serve two very different purposes. When users proactively change the narrative direction, tone, and details, they are buying control over the story; if they are merely correcting character relationships the model got wrong or omitted settings, the same point deduction becomes paying to restore the experience they originally expected.

One reviewer on the Japanese App Store wrote that they had purchased Pass and quite a few points, but the post-update execution of settings and memory performance did not meet expectations; they also stopped using that model after Koji's price increase and said they might not renew. This feedback links the two kinds of spending: users buy membership and points separately, and dissatisfaction with model service can make them reconsider both expenses at once.

The two layers of charging can be separated in the accounts, but not in the experience. Removing ads cannot compensate for an interrupted plot, and model capability cannot replace a clear billing arrangement. The platform needs to distinguish between users actively rewriting and the system failing to complete the original task, and decide which operations are charged and which failures should be retried or compensated. The more granular the charged items, the more specific the responsibility for service delivery becomes.

Free tools expand the room for paid services

To keep premium models continuously useful, there also needs to be content worth users participating in round after round.

In March, Zeta changed its content unit from Character to Plot. A character is just a person in a story, while Plot also accommodates other characters, the environment, and the worldview. At the end of June, four features—Lorebooks, Info Boxes, Style, and User Response Options—were also opened to the free zeta model.

Lorebooks associate worldview materials with keywords and invoke them when the chat involves related content; Info Boxes present story states such as dates, emotions, and items; Style controls narrative perspective, pacing, and expression; response options let users choose the next action. These tools organize the story in the creator's mind into content the model can use.

Input and generation of interactive plots

Opening these tools for free and charging premium models per use can coexist. Creators first build characters and worlds, users enter the content with the basic model, and complex plots then provide opportunities to use premium models. Locking creation tools behind a high price threshold may first limit story supply; opening the tools allows more content to enter the platform, with charges later levied on generation services.

This also adds an operational responsibility to content platforms. Traditional platforms are mainly responsible for making works discoverable, while AI plot platforms are also responsible for making works run while being read. With the same set of settings, changing the model or adjusting the generation strategy may change how the characters and plot are presented. Model updates affect not only the chat feature but also how users experience already accumulated content.

For the platform's investment trade-offs, expanding the number of characters, launching stronger models, and maintaining the usability of old plots need to be arranged together. If a batch of carefully crafted settings becomes difficult to continue using after an update, the platform may lose both creators' investment and users' reasons to keep paying. The richer the content supply, the more important the compatibility work between models and creation tools becomes.

Top-ups have also become product design

Zeta's recent changes have extended to where the money is paid.

In the Japanese web pricing announced on October 29, the monthly Pass subscription is 1,320 yen, versus 1,650 yen in-app; the one-day pass is 240 yen and 300 yen respectively, with both web prices 20% lower. Beyond subscription benefits, the platform has begun using differences in payment channels to rearrange pricing.

Channel prices for Zeta Pass in Japan

The same membership service appearing at different prices on different payment channels shows that the business room is not only on the model side. App stores handle distribution and payment, while web payment provides another transaction path; differences in channel fee structures give the platform room to offer users lower prices. Whether lower prices can also improve platform revenue depends on actual payment fees and transaction arrangements, but lowering user prices does not necessarily require lowering model service standards.

A change closer to the usage process is automatic point top-up. Users first set a balance threshold and the quantity purchased each time; when the balance falls below the threshold, the system automatically purchases and replenishes points, with setup and disabling both done on the web. Actions that originally required exiting the conversation and entering the top-up page are replaced by payment arrangements set in advance.

Automatic point top-up process

This brings subscriptions and point consumption even closer. The monthly fee renews by time, while points trigger purchases by consumption: the former maintains membership benefits, and the latter continues generation services. When the story is not interrupted because the balance runs out, the platform reduces one pause in the consumption process and also reduces the chance for users to reconsider "whether to continue" before each top-up.

Automatic top-up can solve payment interruptions, but whether the model is worth continuing to use is still decided by each generation. For the platform, purchase frequency, point consumption, and story experience must be viewed together: increased consumption may come from more engaged plot advancement, or from repeatedly correcting the same problem, and the two have completely different meanings for subsequent renewals.

Membership benefits, model charging, and web payment have already formed a complete business arrangement for Zeta. To continue running this business, creators' settings must run stably, users' money must be spent on the story they want to happen, and each charge must be enough to support continuous generation. The monthly fee is only the beginning; what follows is round after round of services that need to be delivered.

This article comes from the WeChat public account "Feng Wei Feng Da Bai," author: Feng Da Bai, published by 36Kr with authorization.