ai roi & token cost estimator

Put in how many requests a month, how big they are, and how much of each prompt repeats. You get the monthly API bill per model, the cost of a single task, and what the time saved is worth, so you can see whether the use case pays before anyone writes code.

free, no sign-up, runs in your browseropen source on GitHub

$288API cost per month

$0.0144API cost per call

$90,000value of time saved per month

$89,212net per month after $500 other cost

2,000hours saved per month

114.2×return on running cost

where the API money goes

uncached input	$120.00
cache reads	$6.00
cache writes	$2.25
output	$160.00

same workload on every model

model	per month	per call
Claude Fable 5.2	$1,141	$0.0571
Claude Opus 5.5	$577	$0.0288
Claude Sonnet 5.5	$288	$0.0144
Claude Haiku 5.5	$15	$0.0007
Claude Haiku 5.5 (over 100K)	$74	$0.0037

Claude API list prices per million tokens, checked 2026-11-10: input, output, 5-minute cache write and cache read; Batch API 50% off. Haiku 5.5 rates are for prompts up to 100K tokens. The URL keeps your numbers, so you can share the link.

questions

which prices does it use?

Anthropic's published Claude API list prices per million tokens, including the 5-minute cache-write and cache-read rates. The Batch API takes 50% off every token class. The check date is shown under the table.

what counts as a cached token?

The part of the prompt that is identical across requests, such as a long system prompt, a policy document or tool definitions. Cache reads bill at a small fraction of the input price. Each time the cache expires, the prefix is written again at the cache-write rate.

does it include thinking tokens?

Count them in output tokens. Extended and adaptive thinking bill as output, so a reasoning-heavy step needs a bigger output allowance than its visible answer.

can i share a scenario?

Yes. The URL updates as you type, so a link reopens the same numbers.

more free tools

Want the numbers checked against your own process? Book a 20-minute call.