How much of your AI usage limit is left, and why the app will not say

Short answer: Claude shows its five-hour and weekly limits in its settings, Gemini shows its five-hour and weekly usage in its settings, and ChatGPT mostly tells you only once you hit the limit. The figures exist; the apps rarely put them in front of you.

You are mid-task and the answer stops. "You have reached your limit." No warning, no countdown, and often no clear statement of when it lifts.

Why limits are invisible

Providers meter usage on rolling windows — typically a few hours, and a longer weekly one — and they mostly do not surface the counter. There are real reasons for it. The limit is often not a simple message count but a weighted allowance where a long answer with tool use costs more than a short reply, so any number shown would be approximate and would invite arguments. The thresholds also move; publishing one invites complaints when it changes.

The result is that the one number you actually need — how much is left before you get cut off — is the number you cannot see.

What the platforms do publish

More than they show you. The same endpoints their own interfaces call often carry usage figures that are simply never rendered:

• Some expose a percentage of a window spent, plus the time it resets.
• Some report a plan tier, which is the only reliable statement of which limits apply to you at all.
• Some meter side features — deep research, image generation, voice — separately from the main allowance, which is why you can hit one wall while another is untouched.
• At least one publishes nothing whatsoever, and enforces purely by refusing requests. For that one, any number is a guess.

On the three biggest sites, today:

• Claude has a usage page in its settings that shows your five-hour session and your weekly limit.
• Gemini meters compute in a five-hour window and a weekly limit, and shows both in its settings. More on Gemini's limits.
• ChatGPT mostly tells you when you have reached a limit and when it resets, rather than how close you are.

What you can do without any tools

1. Know your window. Most limits are rolling, not daily. If you were cut off at 2pm on a five-hour window, you recover progressively from 7pm, not at midnight.
2. Front-load the expensive work. Long context, file analysis and tool use consume allowance faster than short questions. Do them at the start of a window.
3. Split across platforms deliberately. If you have access to more than one assistant, put routine work on one and keep the strongest allowance for work that needs it.
4. Watch for the slowdown before the wall. Some providers degrade to a smaller model before refusing outright. If answers suddenly get shorter and worse, you are close.

How Tvara shows it

Tvara reads the allowance figures a platform already publishes to its own interface and puts them in one panel, per account, with the reset time beside each.

Two rules it does not break, because a confident wrong number here is worse than no number:

• A figure nobody published is never invented. If a provider reports nothing, the row says so rather than estimating.
• The window that stops you soonest leads. A weekly figure can look healthier than the five-hour one that is about to cut you off mid-answer, so the row leads with the nearer constraint and names which window it is.

Allowance tracking is free. See the full feature list and its honest limits.