Advanced Claude — Book 7 · The Model Menu and the API

The price sheet, and how to read it. What each of the four models costs per million tokens, and why output is always five times input. How much text a million tokens actually holds, and the tokenizer change that means a lower rate isn't automatically a lower bill. Then the part that pays for the book: prompt caching, which breaks even after a single read, and the Batch API's flat fifty per cent, which stack. It ends with one fully worked example — a real job at $112.75 a month taken to $11.49 without changing the prompt or the model — with every figure computed rather than estimated.