Thinking, Effort, Models and Cost - Claude Cheat Sheet

The model menu and what each costs, the five-level effort dial and how adaptive thinking works, and why length is a separate control. Then cutting the bill: prompt caching, the Batch API, lower effort, and a token trap.

On the sheet:

- The model menu, and what each costs: the four models, their prices and windows

- The effort dial: five levels, from low to max, with high the default

- Thinking is adaptive now: the model calibrates its own reasoning

- Choosing a level: matching effort to three real workloads

- When Claude overthinks: the signs, and how to rein it in

- Guiding the thinking: prompting for reflection and visibility

- Length is a separate dial: why effort is not a length control

- Prompt caching: pay once for a fixed prefix, read it back cheaply

- The Batch API: half price for work nobody is waiting on

- Cutting a bill, in order: cache, batch, lower effort, then a cheaper model

- Tools, extras and a token trap: what search, fetch and code runs cost

Two A4 pages, printed front and back, sized to print once and keep beside you while you work. Written in short sentences and plain English, for readers whose first language is not English and clearer for everyone else because of it.

It pairs with the Thinking, Effort and Depth and The Model Menu and the API books, so you can learn each idea in full and then revise it here in two pages.

Every product fact on this sheet - every plan, price, limit, setting and menu - was checked against Anthropic's own published documentation before the file was made. This is an independent study guide, not affiliated with, authorised by, or endorsed by Anthropic. Instant PDF download. By M.R. Sterling.