Claude Code usage limit reached: what to do

First, read which limit you hit

Claude Code’s error reference lists the messages below. Each one names the limit and its reset time.

On Pro and Max, the allowance is shared with the same account’s Claude chats. Anthropic’s help centre says all activity in both tools counts against the same limits. A long research chat in the morning is budget your afternoon coding session no longer has.

API Error: Server is temporarily limiting requests (not your usage limit) is a different thing. It’s a short throttle on Anthropic’s side, retried automatically since 2.1.199, and none of the steps on this page apply to it.

Check what you have left, before it runs out

/usage shows your plan’s windows and when each one resets. Claude Code can also warn you ahead of time, with a line like You’ve used 85% of your session limit · resets 3:45pm.

A warning is easy to miss mid-task, so if you hit limits often, put the number where you’ll see it. For Pro and Max subscribers, the custom status line receives a rate_limits object with five_hour and seven_day windows. Each window has a used_percentage and a resets_at in Unix seconds.

#!/bin/sh
# ~/.claude/statusline.sh — prints e.g. "5h 62% · 7d 41%"
input=$(cat)
five=$(echo "$input" | jq -r '.rate_limits.five_hour.used_percentage // empty' | cut -d. -f1)
week=$(echo "$input" | jq -r '.rate_limits.seven_day.used_percentage // empty' | cut -d. -f1)
[ -n "$five" ] && printf '5h %s%% · 7d %s%%' "$five" "${week:-?}"

Point statusLine in your settings at the script and make it executable. The fields only appear after the session’s first response, and Claude Code drops a window once its reset time passes, which is why the script tolerates missing values. In the Desktop app, the usage ring next to the model picker shows the same reading.

Wait for the reset without watching the clock

From Claude Code 2.1.234, an interactive session signed in with a claude.ai subscription can wait in place and continue the interrupted task shortly after the reset. While it waits, the bottom of the session reads Usage limit reached · continuing automatically at 3:45pm · esc to cancel. Press Esc at an empty prompt to cancel.

In the CLI, the switch is Continue automatically at usage limit in /config. The Desktop app has its own Auto-continue when limits reset checkbox on the session-limit card, and it’s separate from the CLI setting, so turning one off leaves the other on. The Desktop weekly-limit card doesn’t offer the checkbox.

For a five-hour window, this is often the right answer: queue the task, go to lunch. For a weekly limit it can mean days, so read on.

Switch model family, for Opus and Sonnet limits only

The Opus and Sonnet limits each apply only to requests to that family. When one of them is the message, /model and pick a model from outside the family, and the next turn goes through.

Two things to know. First, this does nothing for session or weekly limits, which Anthropic describes as shared across all models. Second, each model has its own prompt cache, so the first request after switching re-reads the whole conversation with no cache hits. On a long session that first turn is expensive.

Buy usage credits, or raise the spend limit

On Pro and Max, /usage-credits buys additional usage beyond the plan. On Team and Enterprise it sends the request to your admin. Credits are metered against a monthly spend limit. If you hit that ceiling, the message changes to You’ve hit your monthly spend limit, and on Pro and Max you raise it in Settings > Usage on claude.ai.

A permanent upgrade to a higher plan raises the base limits instead. If you hit the weekly cap every week, that’s the honest fix.

The API-key trap

Setting ANTHROPIC_API_KEY is the other documented escape. It’s worth being deliberate about. Once you approve the key in an interactive session, Claude Code bills through that Console key instead of your subscription, even while you’re signed in, and API billing has no hard stop. The account is charged for what it uses.

The trap is forgetting to undo it. An exported key in your shell profile keeps routing every future session through the API long after your subscription window has reset. Anthropic’s own troubleshooting names a stray ANTHROPIC_API_KEY as a cause of unexpected rate-limit errors, too, when it points at a low-tier key. Run /status to see which credential is active, and unset ANTHROPIC_API_KEY to go back to the subscription.

Moving the task to another account you hold

If you legitimately hold a second subscription, such as a personal plan beside a work seat, it has its own meter. Run it from its own configuration directory. The multiple-accounts guide covers the setup and the traps.

What you can’t do is move the stalled session itself. A session’s transcript lives in the projects/ tree of the account that ran it, so claude --resume under the second account doesn’t list it. To continue, start a fresh session on the second account and bring the context with you. A useful handoff has four parts: the original task, the current state of the files, what was tried, and the next step.

Don’t copy transcript files between account directories to force a resume. The receiving account would be continuing a history it didn’t write, and the field note on editing history covers what a transcript has to keep intact to stay resumable.

Doing it in Crowsnest

Crowsnest is ours, so weigh this section accordingly. It changes the last option on this page, because it keeps several Claude Code accounts in view and builds the handoff for you.

The stalled session stays where it was, failed turn included. Quota is never pooled, and the continuation after a failure skips confirmation only when you’ve enabled it and exactly one other account is signed in. Multiple accounts are a Pro feature. The multiple-accounts field note has the measured briefing sizes and every boundary.

Questions behind the search

Related reading