A practical guide for Claude Chat users, Claude Code users, and Claude API users. How to use Claude tokens economically, ecologically, and intelligently.
<aside> 💡
Version 2026-06-09
Author: Paul Van Cotthem - www.turnleaf.be Assistant: Opus 4.8
</aside>

Double-click to enlarge to full size.
This guide is segmented by how you use Claude. Each chapter is self-contained.
Tokens are the unit Claude uses to measure everything — your messages, its replies, uploaded files, and images. Every limit in this guide, every cost, every quota window is denominated in tokens.
The basic unit. A token is roughly one word, or about 4 characters of text. Tokens do not map cleanly to words: punctuation, spaces, and partial words each count separately. Think of tokens as chunks of text — Claude reads and writes in chunks, not character by character.
It goes both ways. Both what you send and what Claude replies are counted. A long reply from Claude costs just as much as a long message from you. And output tokens cost 5× more per token than input tokens — then re-enter the conversation history at input rates on every subsequent turn, making verbose replies expensive twice over.
Images cost tokens too. When you attach an image, Claude converts it into tokens to process it. A typical image costs anywhere from a few hundred to a few thousand tokens depending on size and resolution — often more than several paragraphs of text.
A rough sense of scale: