Claude usage limits explained: the 5-hour window and the weekly cap
Claude plans run on a 5-hour session window and a weekly cap. How both work, where to check what you have left, and how to pace yourself through the week.
Claude has two limits, running on two clocks
At the time of writing (October 2026), paid Claude plans meter your usage against two windows at once:
- A 5-hour session window. It starts with your first message and everything you do in the next five hours draws from the same allowance. When it runs out you wait for that window to end, not for midnight.
- A weekly cap that resets every seven days. It sits on top of the session window: you can have room left in the current session and still be stopped because the week is spent. Some plans also have a separate weekly cap for the most capable models.
Two details catch almost everyone. Chat on claude.ai and Claude Code share the same pool,
so a long coding session eats into what you have left for chat. And starting a new conversation
(or running /clear in Claude Code) gives you a clean context, not a fresh
allowance.
Where to see what you have left
- On the web: in claude.ai, open Settings → Usage. It shows both windows, combined across the web app and Claude Code, with their reset times.
- In Claude Code: run
/usage. Same numbers, without leaving the terminal (it will not run while the agent is in the middle of a task).
Anthropic does not publish exact token quotas, and how fast you burn through them depends a lot on the model and on how much context each message carries. A long session with big files on the most capable model spends many times what a short chat does. So treat those two screens as the source of truth, and anything that quotes a number of messages per plan as an estimate.
The number you actually need is the pace
The usage page tells you how much you have spent. It does not tell you whether you can keep going like this until the reset, and that is the question that matters on a Wednesday. The math is simple: compare the share of the allowance you have used with the share of the week that has gone by.
Say your week resets on Monday morning and it is Thursday morning: three of seven days have gone by, about 43% of the week. If the usage page says 60%, you are spending roughly 20% a day, and the remaining 40% lasts two more days. You run out around Saturday, with two days to go.
Ahead of pace means easing off now (shorter contexts, a lighter model for the easy parts, fewer “just one more” runs), while there is still room to adjust. Behind pace means you can push harder without worrying about the weekend. And if you only work weekdays, count working days, not calendar days, or the math tells you that you have more room than you do.
Let your menu bar do the math
That comparison is what LLM Quota does. You tell it when your plan resets (weekly, daily or a rolling 5-hour window) and roughly how much you have used, and it projects where you will land at the reset: on track, room to spend, or slow down. It can count only working days. It never connects to your Claude account: you type the numbers in, so it works with any capped plan, ChatGPT included, and nothing leaves your machine.
It is a widget inside Webstarted Toolbox, a free menu bar app for macOS and Windows, next to timers, a clipboard history and the rest of the menu bar apps we use every day.
