Free tool
Prompt cache audit
Most cost tools assume prompt cache reads cost 10% of input, for every model. They don't. Enter your numbers and see how far off your tool is — and what fixing your caching is actually worth. Read the write-up behind these numbers →
Cache-read price varies 20x across models — from 2.5% to 50% of input. A tool that treats it as a constant is wrong by up to 4–5x on the line item that caching was meant to shrink.
Choose a model.
We find the rest of the money
This checks one line item. A full diagnosis finds loops, retries, routing waste and attribution gaps — and puts a dollar figure on each. Get a cost diagnosis →