How much does an AI feature cost to run each month?

How much does it cost to run AI in an app?

There are two separate costs. The build cost is what you pay once to have the feature made. The running cost is what the AI provider charges every time your app asks the model to do something.

The running cost is usage-based, so it can be estimated before you build if you state your assumptions.

How does Claude API pricing work?

Tokens are pieces of text that models process, and as a rough estimate one token is about 0.75 of an English word, according to Anthropic's pricing documentation. You are charged for the tokens you send in (input) and the tokens the model writes back (output), and output costs more per token.

Prices are quoted per million tokens. Anthropic also lists a separate, lower price for cache reads. For Sonnet 5.5 that is $0.20 per million tokens.

How much does the Claude API cost: Sonnet 5.5 vs Opus 5.5?

Anthropic's own pages list the following. Sonnet 5.5 is priced the same as Sonnet 5. Opus 5.5 is priced 20% lower per token than Opus 5.

Anthropic also says its tests show Opus 5.5 costs 40% less than Opus 5 on typical workloads at default settings, and that in its own testing Sonnet 5.5 costs up to 30% less per task than Sonnet 5. Both are Anthropic's own claims, not independent results, and they are separate from the per-token prices above.

How much does an AI feature cost per month? A worked example

This is an illustration, not a forecast. The usage numbers below are assumptions chosen only to show the method. They are not facts about any real app. Your numbers will differ, and so will your bill.

First, total the tokens. Input: 3,000 x 1,500 = 4,500,000 tokens, or 4.5 million. Output: 3,000 x 400 = 1,200,000 tokens, or 1.2 million. Then multiply by the listed prices.

With these assumed numbers the model bill is $21 a month on Sonnet 5.5 and $42 on Opus 5.5. This covers only the model. Hosting, database and other services are separate lines, as the app maintenance cost post explains.

One more caution: prices are list prices that can change, so check Anthropic's pricing page for current rates.

How do I estimate the Claude API cost before launch?

Use the same steps as the example, then run them again with higher usage, for example three times your guess, because first guesses can be wrong.

What should a founder do about this?

Ask your builder three things in writing: which model and which tier the feature will use, what they assumed for requests and tokens, and what the monthly bill looks like at low and high usage.

The same applies to AI chatbot development and AI automation projects. At Foundreco, AI feature integration is quoted for each project, and every build comes with a running-cost truth sheet before launch. For the build side, see the cost to build an app with AI guide, or send your idea for a free 24-hour audit.

Sources

Details checked on 6 October 2026.

Common questions about AI running costs