Understanding your usage limits

The five-hour window

This is the large figure in the widget. Session-based usage resets every five hours. It is the one that decides whether you can keep working in the next hour, which is why it gets the foreground and the weekly limit does not.

The window does not start at midnight. It starts with your first request after the last reset: begin at 09:10 and you have until 14:10.

The weekly limit

Max plans also carry a weekly limit that applies across all models. It resets at a fixed time each week assigned to your account — your reset day and time stay the same regardless of when you started using Claude or when the subscription began, and each cycle starts with the full allowance again.

Per-model usage

Models do not consume the same allowance at the same rate. An Opus-tier model eats a given quota considerably faster than Haiku, which is why the widget breaks the rows out separately. When you are running into a limit, switching the mechanical parts of the work to a smaller model often gets you further than waiting does.

See the model matrix

"Resets in"

This refers to the moment the quota refills, not to the widget's next fetch. They are two different clocks and they get confused constantly.

Consumed, or remaining?

The ring shows the consumed share, because that is how both providers report it. People ask the question the other way round, so the full view also spells out what is left.

Where the plan name comes from

The plan shown next to the account name comes from the profile endpoint, not from the local sign-in file. The file names a plan too, but it only learns of a change when the command line next writes it — an account upgraded from Max 5× to Max 20× goes on reading "5x" locally for hours. "Refresh now" asks the profile again.

Faded numbers

When a fetch fails but earlier numbers exist, the widget keeps showing them, faded. A figure whose age is unknown is worse than one that visibly is not current. The tooltip names the reason.

Sign-in

The widget only ever reads the provider command line's sign-in file. No sign-in there, no data to show. It never asks you for credentials itself.

One fetch per account

Several widgets watching the same account share a single fetch. That is a deliberate choice: a status display that consumed the quota it reports on would be a strange piece of engineering.

When a provider changes its format

If a provider changes the shape of its response, the widget can no longer read it, and it shows an error state rather than guessing. That is deliberate — a wrong number is worse than no number.