Verified against Anthropic documentation on · Updated
How much Claude usage have you actually used?
To see how much Claude usage you have left, open Settings › Usage in the Claude web or desktop app, or run /usage in Claude Code. On Pro, Max, Team, and other seat-based plans with included usage, you'll see two meters: a five-hour session limit that resets every five hours, and a weekly usage limit that resets on a schedule assigned to your account. Free accounts have a five-hour session limit that changes with demand.
Click your profile avatar in the lower-left corner and choose Settings › Usage. You'll see your current five-hour session, your weekly limit, the day and time it resets, and any model-specific meters on your account.
Run /usage in any Claude Code session. If you're signed in with your Claude subscription, it shows your five-hour session progress, where you stand on the weekly limit, and whether a rate limit is active.
How Claude usage limits work
The five-hour and weekly limits are separate, with separate resets, but the same work can count toward both. Because the session limit keeps resetting through the day, you can go all week without a session pause and still hit your weekly limit before it comes back.
Five-hour session limit
The session limit paces short-term use and helps manage real-time demand. When you hit it, Claude stops responding until the five-hour window resets. Paid plans generally get a larger session allowance. On Free, it moves up and down with server traffic.
Weekly usage limit
Pro, Max, and Team plans also have an overall weekly limit across all models. It caps your total use over a seven-day cycle, and it resets on a fixed schedule assigned to your account rather than on a calendar week.
Each plan gets a different allowance
Free covers basic chat. Pro, Max 5×, Max 20×, and Team raise both the session and weekly allowances for heavier work, each by a different amount, and the rules aren't identical across plans. Enterprise depends on how the plan is billed.
Usage Credits when you run out
You don't have to wait for a reset. On eligible paid plans, you can turn on Usage Credits in Settings › Usage and keep working, billed at standard API rates.
Claude usage is not one number
Settings › Usage, your Claude Code statusline, and the Anthropic Console can all show a “usage” figure, and each one measures something different. Before you compare two numbers, make sure they're the same kind of usage.
| Kind of usage | What it measures | Where you see it | How it differs |
|---|---|---|---|
| Subscription usage | How much of your plan's included allowance you've used in the current five-hour session and the current week. | Settings › Usage in the Claude web and desktop apps, /usage in Claude Code, and the limit notices Claude shows in the app. | It isn't a per-token dollar charge, and it isn't how full your conversation's context is. At 100%, Claude stops responding until the window resets or Usage Credits take over.Compare plan limits →Where to check → |
| Context window usage | How much of the model's context window the current conversation takes up, out of a window such as 200,000 or 1,000,000 tokens. | The Claude Code statusline (context_window.), the /context command, and long-chat warnings in Claude on the web. | It says nothing about your remaining plan allowance. A conversation can fill 60% of its model context while your weekly subscription allowance remains nearly untouched, and vice versa.Show context in your statusline →Claude Code guide → |
| Claude Code session totals | Claude Code's own running count for one session: tokens used, time elapsed, and what that work would cost at API list prices. | The summaries Claude Code prints after turns, the /usage screen, and the statusline cost field (cost.total_cost_usd). | The dollar figure is a list-price estimate, not your plan percentage. It also won't always match the requests in your local JSONL transcript, because our tests showed Claude Code counting model work that never appears there.What each usage screen shows →Our transcript measurements → |
| API usage and billing | Tokens billed straight to your Anthropic account at published per-million-token rates, paid from a prepaid balance or a card. | The Anthropic Console (console.anthropic.com), or Claude Code when you sign in with an API key (ANTHROPIC_API_KEY). | It's separate from your Pro, Max, or Team subscription. API requests don't use your five-hour or weekly allowance, and your subscription allowance can't pay API invoices.Where API usage shows up →API key or subscription login → |
What we've measured about Claude usage
Anthropic's documentation describes how the limits are meant to work. We run Claude Code sessions under controlled conditions and record what it actually reports: rate-limit events, its own cost ledger, the local transcript, and statusline fields. Each result comes from a small, named sample, and the scope note under it says what it doesn't prove.
Sonnet did the same task for about a fifth of the Opus cost
We ran one Next.js build task 22 times in Claude Code. In the cache-controlled runs, Sonnet 5.5 High's median task cost was $0.0925 against $0.4456 for Opus 5.5 High, about 0.21×. Running the same Opus task after about 205,000 tokens of cached context pushed its median to $1.5157, about 3.4×, and unrelated Vim documentation of the same size did much the same.
Scope: 22 runs of one task in Claude Code 2.1.292, priced at API list rates. These are API-equivalent dollars, not subscription quota. Every build passed the same 22 checks, but the Opus and Sonnet builds weren't identical.
Claude Code counts model work your local transcript doesn't show
In one controlled three-turn session, a cost rebuilt only from the model requests in the main local transcript came in 5.65% below Claude Code's own model-usage total. Claude Code had also counted WebSearch-associated Haiku 4.5 calls and post-turn Opus requests that aren't in that transcript, so a tally built from the transcript alone can come in low.
Scope: The 5.65% gap is from that one session, not a universal discrepancy across sessions or releases.
Where the statusline's cost and context numbers come from
The statusline cost field reads Claude Code's internal session ledger, so it includes that outside-the-transcript work. In a four-turn Opus 5.5 session, the ledger ran about 3.9% higher than a cost rebuilt from the main transcript. The context percentage compares the latest response's context against the window Claude Code resolved for the session, 1,000,000 tokens in this case.
Scope: Claude Code 2.1.289, one four-turn session. It doesn't show that every Opus session gets a 1M window or that 3.9% is typical.
A promotional cloud session left the plan meters at zero
In our October 3 test, Claude Code reported the ccr_promotional rate-limit bucket as active, and its five-hour and seven-day utilization stayed at 0 across four rate-limit events. That fits the documented design, where Cloud Session Credits are used before normal plan usage. We didn't see the promotional balance itself go down.
Scope: One 22-minute Opus 5.5 cloud session. It doesn't tell you what happens once the promotional balance reaches zero or after the credits expire on November 5, 2026 at 08:00 UTC.
The research page has the full test conditions, observation logs, and limitations for all four studies, plus a public JSON manifest of the results.
All Claude usage research →Keep your limits on screen while you work
Settings › Usage and /usage show your limits when you go and look. The open-source claude-usage tracker keeps them in view instead. It runs on your machine, reuses your existing Claude Code login, and shows live five-hour and weekly percentages in your terminal statusline, the macOS menu bar, or a local dashboard.
What do you want to know about your Claude usage?
Every place Claude shows what you have left, on web, mobile, and in Claude Code, compared side by side.
When your five-hour window and your account's weekly limit actually come back, and where to find the exact time.
Session and weekly allowances from Free to Enterprise, plus the rules for continuing past them.
How the window works, what drains it, and how to pace your work around it.
How the seven-day cap works across models and how its reset is scheduled.
Subscription login versus API key billing, what /usage shows, and the context tools.
How to keep working past a limit at standard API rates, including bundle discounts and setup.
Put live five-hour and weekly percentages in Claude Code's statusline.
Frequently asked questions about Claude usage
Why does Claude have both a five-hour session limit and a weekly limit?
The two limits do different jobs. The five-hour session limit paces short bursts of heavy use and resets every five hours. The weekly limit caps your total use over seven days to keep access fair over time. On Pro, Max, and Team, both apply at once, so you can run out of weekly allowance without ever hitting a session pause. Seat-based Enterprise plans can have included limits too. Usage-based Enterprise is billed on consumption instead.
Does Claude Code share usage limits with Claude on the web?
Yes, if you sign in to Claude Code with your Pro, Max, or Team subscription (the OAuth login). Claude Code sessions and browser chats then draw from the same five-hour session limit and weekly allowance. If you sign in with an Anthropic API key instead, you're billed per token through the Anthropic Console, and that usage doesn't count against your subscription limits.
What happens when you reach your Claude usage limit?
Claude stops responding until that limit resets, and Settings › Usage shows when that will be. On eligible paid plans you can keep working right away by turning on Usage Credits there, billed at standard API rates.
What to do when you hit a limit →When Claude usage resets →How Usage Credits work →Limits by plan →