Inside Claude Code, run /usage: on a Pro, Max, Team, or Enterprise plan it draws progress bars for your 5-hour session window and your weekly limits. On claude.ai, go to Settings then Usage for the same bars. On an API key, read the anthropic-ratelimit-* headers from any response, remembering those are per-minute throughput limits rather than the subscription windows. Session-file tools such as ccusage show consumption only: remaining allowance is never written to disk.
- Remaining allowance lives in server responses, never in a local file.
/usageis the answer in the CLI./costis now an alias for it.- On claude.ai the same bars are under Settings, then Usage.
- Two windows: a rolling 5-hour session limit and weekly limits.
- Weekly is split: Opus only, and all other models. Check both.
The fast answers
Where to look, by surface. Verified August 2026.
| Surface | How | Shows |
|---|---|---|
| Claude Code | /usage | Progress bars for the 5-hour and weekly windows |
| claude.ai | Settings, then Usage | The same bars, plus reset times |
| VS Code extension | The Account and usage dialog | The same breakdown, Day and Week toggle |
| API key | anthropic-ratelimit-* response headers | Per-minute throughput remaining and reset |
| Menu bar, phone, or watch | A monitor that reads those headers | Live and continuous, no command |
Inside Claude Code
Run /usage
/usageOn a Pro, Max, Team, or Enterprise plan this renders plan usage bars for your 5-hour session window and your weekly limits, alongside activity stats. /cost is an alias for the same command as of August 2026, so there is no separate cost view to learn.
Press d or w to change the window
The breakdown underneath the bars covers either the last 24 hours or the last 7 days. It attributes recent usage to skills, subagents, plugins, and individual MCP servers as a percentage of the total, and flags any behaviour accounting for 10 percent or more, such as long context or cache misses.
Read the Session block for tokens, not for billing
Total cost: $0.55
Total duration (API): 6m 20s
Total duration (wall): 6h 33m 10s
Usage by model:
claude-sonnet-4-6: 1.2k input, 5.3k output, 940.0k cache read, 50.0k cache write ($0.55)That dollar figure is computed locally from token counts at standard list rates. It ignores promotional pricing and contracted discounts, and on a subscription no money moved at all. It is a measure of intensity, not a bill.
The windows you are being measured against
How the subscription limits behave.
| Window | Resets | Check when | Warning sign |
|---|---|---|---|
| 5-hour session | Rolling, five hours after you start using it | Before starting anything long | Under 20 percent |
| Weekly, all other models | A fixed time each week assigned to your account | Midweek, deliberately | Under 30 percent before Wednesday |
| Weekly, Opus only | Same fixed weekly time | Whenever you switch to Opus | Any Opus-heavy day |
The weekly cap is the one people are surprised by, because it is the only limit you can hit without any individual day looking unusual. Five ordinary days in a row will do it, and the first sign is being blocked on a Thursday for no visible reason.
Reading it from the API directly
If you have an API key, the headers are the source of truth and a single cheap request will show them.
curl -sS -D - -o /dev/null https://api.anthropic.com/v1/messages \
-H "x-api-key: $ANTHROPIC_API_KEY" \
-H "anthropic-version: 2023-06-01" \
-H "content-type: application/json" \
-d '{"model":"claude-haiku-4-5","max_tokens":1,
"messages":[{"role":"user","content":"hi"}]}' \
| grep -i -E 'ratelimit|retry-after'The headers Anthropic returns.
| Header | Meaning |
|---|---|
anthropic-ratelimit-requests-limit / -remaining | Requests allowed and left in the current window |
anthropic-ratelimit-input-tokens-remaining | Input tokens left, rounded to the nearest thousand |
anthropic-ratelimit-output-tokens-remaining | Output tokens left, rounded to the nearest thousand |
anthropic-ratelimit-tokens-remaining | The most restrictive token limit currently in effect |
anthropic-ratelimit-*-reset | When that limit is fully replenished, in RFC 3339 |
retry-after | Present on a 429. Seconds to wait; earlier retries fail |
What to do at each level
- Above 50 percent: nothing. Do the work.
- 20 to 50 percent: do not start a long autonomous run you cannot finish.
- Under 20 percent: switch to a cheaper model, finish the current task, commit.
- Under 5 percent: commit now. Getting cut off mid-edit is the one genuinely costly outcome.
- Weekly under 30 percent on a Tuesday: change something structural, not tactically.
The habit worth building is committing before a window closes rather than after. An agent interrupted mid-edit leaves a working tree you then have to reason about, and that is the real cost of a rate limit.
Not having to check
Every method above requires you to stop and ask, and the moment you most need the answer is the moment you are least likely to.
Questions people ask
Run /usage inside Claude Code for progress bars against your 5-hour and weekly limits. On claude.ai, open Settings then Usage for the same bars. On an API key, read the anthropic-ratelimit headers from any response.
It still works and is now an alias for /usage. The per-session token and dollar figures appear in the Session block at the top of the /usage screen, and they reset when /clear starts a new session.
No. It parses session files, which record what you consumed. Remaining allowance is only reported by the server, in rate-limit response headers and the usage endpoint the CLI queries.
Not as absolute token counts. How far an allowance goes depends on the model, the effort level, and how much context each turn carries, so a single number would be misleading. The progress bars are the published answer.
The plan usage bars come from the server, but the attribution breakdown beneath them is computed from local session history on that machine only. Work done on another laptop or in claude.ai is not included in the breakdown.
Not for the session or weekly windows, which are shared across models. It does help after the model-specific "You have hit your Opus limit" message, because the Opus weekly limit is separate.
No. They report per-minute throughput limits in requests, input tokens, and output tokens, replenished continuously. The 5-hour and weekly subscription windows are a different measurement and the two will not agree.
Not from Anthropic directly. A monitor that reads the rate-limit signal can warn at a threshold you set; Continuum does this in the menu bar and on your phone, with a burn rate so the warning is expressed as time remaining.
Sources
Every figure above was read from these pages on August 2026. Vendors reprice without notice; if you find a stale number, tell us.