Official Documentation: Costs, Usage and Models

Two reference pages worth opening once, and worth coming back to whenever a figure you read somewhere does not match what you are seeing.

Costs and usage

• https://docs.claude.com/en/docs/claude-code/costs

What gets counted and how the caching discount works. This is where you check which setup a number describes, because the figures people repeat online often belong to somebody else's.

The clearest example is the caching timer. You will see five minutes quoted for it, often, and it is a real number. Claude Code picks the timer based on how you signed in. On a subscription it asks for the hour. On an API key, or through one of the cloud platforms, five minutes is the default, and that is where the figure people repeat comes from. It is true, and it is not describing your setup.

Model configuration

• https://docs.claude.com/en/docs/claude-code/model-config

The model names shift over time, so this is the page to check rather than a lecture. It also covers setting a model for the smaller helpers Claude sends off while it works, which is one of the highest-value adjustments you can make, because those helpers quietly inherit the expensive model unless you tell them otherwise.

Two things to carry into both pages

The figure is a measure of work, not a bill. If you are on a subscription, what /usage shows you is what the same work would have cost at pay-as-you-go rates. Read it as a way to compare one session or one choice against another.

Switching models mid-session has a cost. Claude has to re-read the room for the new model, so that discount gets rebuilt from scratch. Switching is useful. Flipping back and forth constantly means paying that rebuild every time.

---

This resource is part of

Claude Code: From Beginner to Pro Agentic AI Engineer

by Arun Nagarathanam

Take your skills to the next level with

AI GEO 2026: Get Recommended by ChatGPT, Google, Perplexity

by Arun Nagarathanam