Cost & limits

The written-up reasoning behind this site's calculators: why your bill doesn't match a naive token count, when the subscription beats the API and when it doesn't, how caching and batching actually move the number, and what changes the moment an introductory price expires. If you want the computed answer for your own numbers, use the tools; these guides are for understanding why the tools compute what they compute.

Cost on the Claude API is rarely a single number you can quote from memory, because it's a function of at least four things that move independently: which model you're calling, how much of your input is cacheable, whether the request can tolerate batch latency, and which effort level it's running at. Most of the guides in this cluster take one of those four variables and work through it in isolation with real figures, rather than trying to explain the whole cost model in one sweep — that tends to be more useful when you're trying to fix a specific line item on a bill that looks wrong, rather than trying to understand billing in the abstract.

If a number here doesn't match what you're seeing in your own usage dashboard, the usage tracker and reading a usage block correctly guide are the fastest way to find where the discrepancy actually is.