Quick answer
Claude Code writes token counts for every request to local session files. /usage summarizes the last day or week on that computer. UsageAtlas reads the same files and shows tokens by day, project, session, and model, with input, output, and cache tokens split out.
Where the numbers come from
Claude Code saves each session as a file under .claude/projects in your home folder. Every reply records its input, output, cache write, and cache read tokens. Any tracker that shows Claude Code tokens reads these files.
See them in UsageAtlas
- Install UsageAtlas. It finds Claude Code on its own.
- Open History and pick 7, 30, or 90 days.
- Filter to Claude, then break the total down by model, project, or session.
How to read the totals
Expect cache reads to dwarf everything else. Claude Code resends the conversation on each request, and the cached part counts again every time at a lower price. Output tokens are the smallest number and the most expensive per token. The token types guide explains each one.
Questions, answered
Why is my Claude Code token count so high?
Most of it is cache reads. Claude Code resends the conversation with each request, and cached tokens count every time at a much lower price.
Can I see Claude Code tokens per model?
/usage shows recent activity. UsageAtlas splits tokens by model across 7, 30, or 90 days.
Does it include usage from my other computer?
Token history comes from local files, so each computer has its own. The limit percentages cover your whole account.