agent cost calculator · free tool
What does your agent really cost?
Every turn re-sends the whole context. Enter your numbers to see the bill with and without prompt caching.
per month
–without caching
–
with caching
–
–
cost of one conversation, turn by turn
| Input tokens sent | – |
|---|---|
| …of which cache reads | – |
| Cost without caching | – |
| Cost with caching | – |
Your real numbers will differ.
Cache misses, model choice and prompt structure decide what you actually pay. A production audit measures it on your real traffic.
How this is calculated
Turn t sends the stable prefix, every earlier turn (input + output), and the new input. Without caching, all of it is billed at the input price.
With caching, everything sent in the previous request is a cache read; only the tokens added since then are written to cache. The first turn writes the whole prefix.
It assumes the cache stays warm between turns. Long pauses can expire the cache and turn reads back into writes, so treat this as a best case.