Claude Code's weekly limit, and why it catches people out

The 5-hour window is an inconvenience - you make coffee and it comes back. The weekly cap is the one that takes the rest of your week. And most of what is written about it online is confidently wrong, including several pages that currently rank above this one.

So: what Anthropic actually documents, what it deliberately does not, and how to find the only number that matters - yours.

Start here: nobody on the internet knows your reset time

The most repeated claim about the weekly limit is that it is a rolling seven-day window that resets seven days after you started using it. Treat that with suspicion. Anthropic's help centre describes the weekly reset as happening at a time tied to your account, shown to you directly - not as a countdown from your own first message.

You do not have to take my word for it either, and you should not. Two places show you your actual figure:

• Claude → Settings → Usage. Progress bars for your 5-hour session and your weekly limits, and the reset times that go with them.
• Type /status in Claude Code. Your remaining allocation, in the terminal you are already in.

That is the authority. This page, every other blog post, and every usage tracker including mine are downstream of it.

There is more than one weekly limit

This is the detail that produces most of the confusion. Anthropic's documentation describes weekly resets for Opus separately from all other models. In practice that means you can be completely out of Opus for four days while still having plenty of Sonnet capacity - or the reverse.

If your tracker shows you one weekly number and you are surprised by a refusal, that is usually why: the number you were watching was not the limit you hit.

The other structural detail people miss: Claude and Claude Code share one pool. Anthropic states that all activity in both tools counts against the same usage limits. An afternoon of chatting on claude.ai comes out of the same weekly budget as your agent runs - which is also why a purely local usage tracker can undercount, since it can only see what happened on that machine.

Why the exact number is never published

Anthropic's own support article says plainly that exact token limits are not published. That is a deliberate product decision, not an oversight, and it has one big consequence for you:

This is worth knowing before you trust any percentage, mine included. MeterTab handles it by using your account's own official percentage when you connect it, and by learning your real limit from how far your local token counts move that official number - so the gauge stays calibrated afterwards, including on your phone and watch. When some of your usage happened somewhere it cannot see, it says so rather than presenting a confident wrong figure.

The five things that actually burn the week

Almost every "I burned my weekly limit by Tuesday" story traces back to one of these, and none of them are exotic:

1. Not clearing context. Every previous message is resent on every turn. A session that has wandered through three unrelated problems pays for all three, on every single new message. /clear between tasks, /compact within a long one. This is the single biggest one.
2. A bloated project notes file. Your CLAUDE.md is prepended to everything. Keep it under roughly 200 lines and delete anything no longer true.
3. Leaving Opus on by default. Opus costs several times more per turn than Sonnet, and has its own weekly ceiling. Switch up when the task needs it, not as a standing setting.
4. Executing before planning. For anything touching more than two or three files, use Plan Mode - or just ask it to list the files it will touch first. A wrong direction discovered after twenty edits costs you all twenty.
5. Unwatched long runs. An overnight agent, a big parallel refactor, a loop that retries a failing command forty times. This is the one that turns a bad afternoon into a lost week, because nobody was looking.

When you have already hit it

There is no trick that resets the counter early. Anthropic's options are: wait for the reset, turn on usage credits to keep going at standard API rates, move to pay-as-you-go billing through the Console, or upgrade the plan. Anything on the internet promising otherwise is selling you something or reselling API access.

The practical move is to make the weekly cap visible before it becomes a decision. Two things help more than anything else:

• Something that shows the tighter of your two windows without being asked. Checking /status works, but only when you remember, and you remember least on the days it matters most.
• Something that stops a run rather than warning about it. A warning cannot be seen by an agent. A Claude Code PreToolUse hook can refuse the tool call outright - here is how to build one, free, in about fifteen lines.

Related

• When does the 5-hour window actually reset?
• Stop Claude Code before it hits your limit
• Is Claude down, or am I out of tokens?

Checked against Anthropic's public help centre on 3 August 2026. Limits and plan structures change - your own Settings → Usage page is always more current than this one.