How to check Claude Code token usage
Start with /usage in the current Claude Code interface. Its session block shows token statistics and an estimated cost; subscription accounts can also show plan usage. For the billed total, consult the account's billing console. A session figure is not an invoice.
Step 1: Identify your setup
Run /status to check the active setup and /model to check the model. Determine whether the work uses a subscription, the direct API or a cloud provider. Your measurement should answer a specific question: remaining allowance, current-session activity or billed spending.
Step 2: Open the right view
| View | Use it for |
|---|---|
| /usage | Session information and available plan usage |
| /context | Material currently loaded in the conversation |
| /status and /model | Active setup and selected model |
| Account billing console | Authoritative billed spending |
| /cost on older versions | Session cost where the installed help supports it |
Current documentation centers on /usage; /cost appears in older versions and some support material. Check your installed help when a command differs. Avoid comparing measurements from different billing setups as though they have the same meaning.
Step 3: Read the session quantities
Session quantities can distinguish fresh input, generated output and cached material. Cost estimates multiply those categories by rates. A subscription's displayed session dollar figure does not mean you owe that amount on top of the plan.
Context occupancy answers another question: how much material is currently loaded. It does not show how much account allowance remains. Read the limits guide when the interface reports a restriction.
Step 4: Build your own baseline
Use a small worksheet with these columns:
Task | Model | Billing setup | Session quantities | Displayed allowance | Notes
Record a few tasks you do regularly. Note whether a task used large files, lengthy outputs or repeated requests. Keep private prompts and source code out of any shared worksheet. Compare similar tasks before drawing conclusions about improvements.
For an API scenario, put observed quantities into the Claude token cost calculator. Its monthly estimate repeats a request, so it is useful for a known repeat workflow. A varied coding month needs multiple scenarios or actual usage records.
How to interpret a change
If two runs differ, inspect the model and the material each run processed before assuming a price or allowance changed. Shorter visible prompts do not necessarily mean less total session usage. Task history and tool activity can differ substantially.
This site provides local counting and manual budgeting. It does not connect to your Claude account or collect usage automatically. For a local text sample, use the token counter; for exact account measurements, use the account's own reporting.
Sources & review
Checked 3 October 2026. Product rules can change; consult the linked official references.