A tab’s tokens
Each tab counts the tokens it has used. The count sits in the composer bar, at its right, in short form such as 1.2M tok. It moves while Claude works, and it’s saved with the tab, so it survives restarts and resumed sessions.
The breakdown
Click the count (its tip says Tokens used by this tab) to open a breakdown:
- A table with a row for each model the tab used: Input, Output, Cache write, Cache read and Cost.
- A summary line: 12 turns · 345,000 tokens · about $1.20 at list price (an estimate, not your bill). Claude Code works out the cost at list price, so it’s an estimate, not what you’re billed.
- This session window and a number: the tokens this tab has used since the current 5-hour window started. That’s the part that counts against the session limit.
- Tokens per turn, a small chart of the tab’s last turns from the past week.
Each turn adds what it used, subagents and compaction included, so a long session isn’t counted again with every turn.
The tab’s info card (hover over its row, or the ⓘ button at the end of the line under the message box) has a Tokens row with the same count and the number of turns.
Where else the numbers show
- The working line above the composer shows the tokens of the turn in progress: see The working line and running tasks.
- The Usage window and the detailed header rank your tabs by what they used in the session window.
- The count isn’t the same as how full the tab’s context is: see Context and compacting.
Resetting the counts
Clear usage history… in Settings → Usage can also reset every tab’s token totals, with Also reset per-tab token totals. Otherwise the totals stay with the tab: see Usage history.
See also
- The composer bar for everything else in it.
- The prompt cache, which is why cache read tokens cost so much less.