How to Price an AI App: Subscriptions, Credits, or One-Time

Pricing an AI app is harder than pricing any other kind of software, because every user costs you a different amount. One person sends five messages a month. Another runs a thousand image generations a week. If you charge them both the same flat price, one subsidizes the other, and it is usually the one who costs you the most.

The good news is that the choice is not as open as it looks. For most AI apps, the right answer is a hybrid: a base subscription with a defined usage allowance, and overage billing for customers who go beyond it (Freemius). This guide walks through the three pure models, where each one breaks, and how to build a hybrid that protects your margin.

The three pricing models

Model	How it works	Best for
Subscription	Flat monthly or annual fee, unlimited or capped usage	Apps with predictable, low-variance usage
Credits (prepaid)	Users buy a balance of credits and consume against each action	Apps with high usage variance and clear per-action cost
One-time purchase	Single payment, usually with a token or usage budget baked in	Tools and utilities, not ongoing AI services

These are not religious commitments. Most successful AI products combine two of them. The decision is which one carries the base price and which one handles the variance.

Why flat subscriptions break for AI

A flat subscription hides the cost variance between users, and that variance is exactly what makes AI apps expensive to run. A token is the unit of work an AI model charges for, roughly 0.75 words or four characters, and every token costs you money (AI Pricing Guru, MindStudio).

The failure mode is predictable: a small number of power users consume a disproportionate share of your inference cost, and a flat price means the heavy users are undercharged while light users are overcharged. The fix is not to raise the flat price, because that chases away the light users who make the product feel alive. The fix is to make the price reflect usage.

Credits do this directly. They expose the usage variance and make the system transparent: a user buys a balance, consumes against it, and adds more when they run out (Vellum). The downside is friction: a user who has to think about credits before every action will use the product less.

The hybrid most AI apps should use

The pattern that works for most AI apps is a base subscription plus usage allowance plus overage. A user pays a monthly fee that includes a fair amount of usage, and if they go beyond it, they either buy more credits or pay a metered overage rate (Freemius).

This structure solves three problems at once. The subscription gives you predictable revenue and a number to optimize churn against. The usage allowance protects you from power users who would otherwise cost more than they pay. And the overage gives heavy users a clear, fair path to keep using the app instead of being silently capped.

To set the numbers, start by modeling your per-user inference cost. Estimate how many tokens or generations an average user consumes in a month, multiply by your model cost, and set the subscription price at roughly three to five times that figure. The multiple matters: inference cost is not your only cost, and the margin has to cover support, refunds, store commission, and the power users who come in above average (Causo).

One-time purchases, and when they make sense

A one-time purchase is the wrong default for an AI app, because it caps your revenue while your costs keep running. But it has a place: for a tool that does a bounded job, like a one-time document analysis or a fixed set of generations, a single payment with a clear usage budget can work better than a subscription no one wants.

If you do offer a one-time tier, be explicit about what is included. A user who buys lifetime AI access for $29 will cost you money within months if they actually use it. Price the one-time as a usage allowance, not as unlimited access, and make the cap obvious in the purchase copy.

The commission problem

Whatever price you set, Apple and Google take a cut. Both stores charge a commission on digital purchases: 30% in the first year of a subscription and 15% thereafter on the App Store, with Google Play taking 15% to 30% depending on revenue (Groovyweb).

This matters most for AI apps because the commission applies to the gross price, not to your margin. If a user pays $10 and your inference cost is $7, the store takes $3 and you are left with nothing. For any AI app where usage cost is a large share of revenue, the hybrid model is not just nice to have; it is the only way to keep the store commission from eating your entire margin.

One practical note: consumable credits sold as In-App Purchases are subject to the same commission. There is no store-approved way around this for digital goods consumed inside the app. Plan your pricing with the commission baked in from the start.

A simple framework for choosing

Estimate per-user inference cost for a typical month. If you cannot estimate it, you are not ready to price yet. Measure it first.
Decide on the base model. Predictable, low-variance usage leans subscription. High variance with clear per-action cost leans credits.
Add the other model for the variance. Subscription plus overage credits, or credits plus an optional subscription for heavy users.
Set the price at 3 to 5 times per-user inference cost, then confirm it clears the store commission and leaves a real margin.
Test the paywall before you ship. The price is only half the work. The paywall has to sell the value, not just state the number.

If your app is a Flutter app, the paywall implementation is its own problem, and the place most teams get it wrong is restore purchases and entitlement handling rather than pricing. Our companion guide on secret management and the broader paywall analyzer cover the implementation side.

The short version

Flat subscriptions fail for AI apps because a few power users cost more than they pay. Credits fix the variance but add friction. The answer for most apps is a hybrid: a subscription with a usage allowance and metered overage, priced at three to five times your per-user inference cost, with the 15% to 30% store commission built in from the start.

If you want to model the numbers for your own app, use the free monetization calculator to see what a subscription, credit, or hybrid structure does to your margin at different usage levels.

Sources

Frequently asked questions

How much should I charge for my AI app?

AI app pricing depends on your per-user API costs, target audience, and value delivered. Common models range from $5-15/month for consumer AI apps to $20-50/month for prosumer tools. Always ensure your subscription price covers your API costs with margin. Start higher than you think — you can always lower.

Should I offer a free tier for my AI app?

A free tier with limited usage helps users try before they buy, but with AI apps, API costs mean every free user costs you money. Use a generous free trial (7-14 days) instead of a permanent free tier, or limit free users to 3-5 AI generations per day.

What subscription models work best for AI apps?

The most successful AI apps use tiered subscriptions: a free trial (7-14 days), a basic plan ($5-15/month) with usage limits, and a pro plan ($20-50/month) with higher limits. Server-side entitlement validation is critical to prevent bypassing paywalls.