I hit my Claude limit - here's what to do in the next 5 minutes

Hitting your Claude limit costs you minutes, not your work. Do four things: read your real reset time on the banner or in Settings > Usage, leave everything open because your context survives, decide between waiting, switching, and upgrading based on how far away the reset is, and set up a warning so the next cap shows up in your peripheral vision instead of mid-sentence.

Update, 2026-09-19: Max plans no longer run a separate weekly limit scoped to Opus. Opus and Sonnet are both current, fully available models, and both draw from the single weekly limit that covers every model. The second weekly meter on Max is scoped to Fable instead: Max includes Fable for up to 50% of the weekly limit, after which Fable keeps running on pay-as-you-go usage credits. Pro has one weekly limit across all models and bills Fable from usage credits from the first message. Checked against a live Claude Max 5x account that used Fable, Opus and Sonnet in the same week: Settings > Usage shows current session, this week, and Fable this week, with no Opus meter.

The short answer

Minute 1: find out which cap hit you and when it resets

Claude runs three rolling caps, and they lock you out for very different lengths of time. The first thing to do is identify which one fired, because everything else in this plan depends on whether you are waiting 20 minutes or 4 days.

The banner that blocked your message names the cap and shows the reset time. For the full picture, open Settings > Usage on claude.ai - per Anthropic's help-center article on usage limits, that page is the authoritative view, with each cap's percentage and its exact reset time in your local timezone. In Claude Code, the /usage command prints the same numbers in your terminal, and a menu-bar meter like tokn.watch shows the same percentages live, which is how you see the next one coming before it lands.

One distinction worth 10 seconds: if you got a "Claude is at capacity" error rather than a limit banner, you did not hit your cap at all. That is peak-hours congestion on Anthropic's side, it costs you nothing, and the fix is to retry in a few minutes or shift the session off the 9 AM-5 PM US Pacific block. The rest of this article is about real limit hits. The mechanics of all three caps are covered in depth in how Claude's usage limits really work in 2026.

Minute 2: protect your context - you lost budget, not work

Nothing was deleted. The conversation stays in your sidebar, Claude Code keeps your session on disk, and every file Claude touched is exactly where it was. When the window completes, you resume from the same point with the full context intact.

So the move in minute 2 is to not do the destructive thing. The classic frustration responses - closing the conversation, killing the terminal, deciding to "start clean later" - are the only ways to actually lose something here. Re-explaining a complex task from scratch after the reset burns more of your fresh budget than the limit hit cost you, because the most expensive part of any Claude session is rebuilding context. The maths of that is the same reason long conversations get expensive, covered in 13 ways to save Claude tokens.

Two 30-second actions that pay off after the reset:

• Write down the next step while it is fresh. One sentence in a notes file: "resume by running the failing test in auth_test, the fix is half-applied in session.ts". Future you resumes in seconds instead of re-reading the whole thread.
• Note your reset time somewhere you will see it. The banner shows it once; if you dismiss it and walk away, you are back to guessing. Settings > Usage always has it.

Minutes 3-4: decide - wait, switch, or upgrade

This is the only real decision in the 5 minutes, and it is determined almost entirely by the reset time you found in minute 1.

If you wait: make the lockout productive

A limit hit is a forced review break, and most coding sessions are overdue for one. The highest-value work while locked out is exactly the work Claude was doing badly anyway because you were moving too fast: read the diff Claude produced, run the tests yourself, write the spec for the next task, clean up the branch. Per Anthropic's usage-limit best practices, batching your asks and arriving with a tight plan is also how you make the refreshed budget last longer.

If you switch: models, providers, credits

• Model switch only rescues an Opus-cap hit. The 5-hour session cap and the weekly all-models cap are shared pools - every model draws from them, so dropping to Sonnet or Haiku after those caps fire does nothing. After an Opus-cap hit on Max, though, switching to Sonnet genuinely keeps the session alive.
• Provider switch works because the budgets are unrelated. Your ChatGPT limits are a separate system on separate timers. If you run both subscriptions, hitting one cap just moves the work to the other tool - the trade-offs are mapped in Claude vs ChatGPT limits, side by side.
• Extra-usage credits are the Max-plan escape hatch. They unlock as a parallel budget once the weekly cap is exhausted. They do not reset the cap; they let you keep working past it at pay-as-you-go rates.

If you upgrade: do it on evidence, not frustration

One limit hit during an unusually long day is normal Pro usage. Hitting the cap every week is the actual upgrade signal. Before paying 5x for Max 5x, look at your last month: if the hits cluster around rare crunch days, credits or pacing beat a plan change; if they are weekly, the decision framework in Claude Pro vs Max will settle the tier in 10 minutes. The pricing is on Anthropic's pricing page.

If you are in Claude Code when it hits

The same caps fire, with two extra wrinkles. First, Claude Code burns the shared pool faster than chat does - every tool call, file read, and test run costs tokens, which is why the 5-hour cap disproportionately bites CLI users mid-task (a dynamic The Register covered when the weekly caps landed). Second, the lockout is account-wide: the chat interface is locked too, because claude.ai, Claude Code, and the apps all draw from one pool.

The Claude Code specifics:

• Your session survives on disk. When the window completes, claude --resume (or --continue) picks up the same conversation with the same context. Do not delete the worktree in frustration - half-applied changes are sitting there waiting.
• /usage is your in-terminal dashboard. Same numbers as Settings > Usage, zero context switch.
• After the reset, downshift the model. If the hit was Opus-related, /model to Sonnet keeps the next window alive several times longer for most coding tasks.
• Commit before you walk away. The lockout is the natural commit point - you want the half-done state recoverable even if the session file is not.

What will not unlock it (save yourself the 10 minutes)

Every one of these gets tried in the first minutes of a lockout, and none of them work, because the cap is server-side and account-level:

• Refreshing, logging out and back in. The banner re-renders; the cap stays.
• Starting a new conversation. Caps are account-level, not conversation-level.
• Switching surfaces. claude.ai, Claude Code, desktop, and mobile share the same pool.
• Switching models - unless it was specifically Fable's 50% ceiling on Max, as above. Sonnet, Haiku and Opus all draw from the same weekly pool, so swapping between them rescues nothing.
• Waiting for midnight or Sunday. Claude's windows are rolling, anchored on your own first message of the cycle. The reset time on the banner is the real one - the full mechanics are in when does Claude reset.

Minute 5: make sure the next one never blindsides you

The lockout you just hit was visible coming for hours - the percentage was sitting in Settings > Usage the whole time. The problem is that nobody loads a settings page every 20 minutes while deep in work. The fix is to move the number somewhere you cannot avoid seeing it:

• Check Settings > Usage at session start. Free, takes 5 seconds, tells you whether today is a "burn freely" day or a "pace yourself" day.
• Run /usage at natural breakpoints in Claude Code - after a big refactor, before starting the next one.
• Put the percentage in your menu bar. A meter that is always on screen turns the limit from an ambush into a fuel gauge you glance at. That is the entire reason tokn.watch exists.
• Spend the budget deliberately. Sonnet by default, new chats for new topics, tight prompts - the pacing tactics that stretch each window are in the token-saving guide, and the full survey of tracking options is in the complete guide to tracking your Claude usage.

Frequently asked questions