Models and costs

Everything AI on Maypop is paid for in credits: building in Studio, generating media, and AI features inside apps. This page explains where credits go, so you can pick the right model and make them last.

The tiers

You choose a tier, not a specific model. Maypop picks a strong model for each tier and upgrades them over time.

Tier	Use it for	In Studio	In apps (maypop.ai)
Fast	Small, clear changes: copy, colours, a new button	All plans	fast
Smart	Most work: new apps, features, fixing bugs. The default.	All plans	smart
Pro	Hard problems: tricky logic, big restructures, stubborn bugs	Pro plan	—

Higher tiers think harder and cost more per step. A good habit is to build with Smart, drop to Fast for polish, and reach for Pro when Smart gets stuck.

How text models are charged

AI models read and write tokens — pieces of words, roughly four characters of English each. Every request is charged for three kinds:

• Input — what the model reads: your message, the conversation so far, and, in Studio, the parts of your app it’s looking at.
• Cached input — input the model read moments ago and can reuse. Much cheaper; see below.
• Output — what the model writes: its reply and the code it changes.

Credits per million tokens:

Tier	Input	Cached input	Output
Fast	20	2	100
Smart	400	20	2,000
Pro	800	40	4,000

Apps calling maypop.ai with fast or smart pay the same rates as Studio.

A credit is worth the same whichever tier spends it, so tiers compare directly: a Smart step costs about twenty times a Fast one, and Pro about twice Smart.

Caching: why long sessions cost less than they look

Each time you send a message in Studio, the agent works in steps — read a file, edit it, run the app, check the result — and every step sends the whole conversation back to the model. Without caching, a long session would pay for the same text over and over.

With caching, the part of the input the model just saw is billed at the cached rate: a tenth of the normal input price on Fast, a twentieth on Smart and Pro. Only the new part — your latest message, the result of the last step — pays full price.

Caches expire after a short idle time, so a burst of changes in one sitting caches better than the same changes spread across a day.

A worked example

One Smart step in a running Studio session that reads 50,000 tokens, of which 45,000 were cached from the step before, and writes 2,000:

Part	Tokens	Rate (per million)	Credits
Fresh input	5,000	400	2.0
Cached input	45,000	20	0.9
Output	2,000	2,000	4.0
Step total			6.9

Without the cache, the same step would cost 24 credits. One message to the agent usually takes several steps, so a change costs a few steps’ worth; Usage and billing in your settings shows what you actually spent.

Decisions

maypop.ai.decide — classifying, routing, yes/no checks — is priced on input only: 8.4 credits per million input tokens, at least 0.01 credits a call. A decision over a short message costs that minimum, so one credit buys about a hundred of them.

Images, audio and video

Media is priced per result, the same in Studio, in apps, and with maypop ai:

What	Credits
Image	6 per image
Audio	1 per second generated (up to 120)
Video, fast	15 per second (4–15 seconds)
Video, quality	30 per second (4–30 seconds)
Transcription	about 1.3 per MB of audio, at least 0.25

An 8-second fast video is 120 credits; the same in quality is 240.

How a charge happens

1. Before a request runs, Maypop sets aside an estimate so you can’t go past your limit mid-request.
2. When it finishes, you’re charged what it really used, and the rest of the estimate is released.
3. A request that fails costs nothing.

When you run out, AI requests stop with a clear message until your daily allowance resets, your monthly extra renews, or you top up — see Plans and credits.

Who pays

• Studio and maypop ai: you, the person building.
• AI inside an app: the signed-in person using the app, not its creator. Sharing an AI app with a hundred people costs you nothing. Anonymous visitors can’t use AI until they sign in.
• Builds from an AI assistant over MCP: you, at the same Studio rates.

Spend less

• Say what you want in one clear message rather than many small nudges.
• Use Select element to point at the part you mean, so the agent reads less of your app.
• Use Fast for wording, colours and spacing.
• Batch related changes in one sitting, while the cache is warm.
• In your own app, pick fast for simple tasks like tagging or summarising, and keep smart for reasoning.