Claude too many requests error (429): Anthropic's throttle, not your plan limit, and the six fixes in order

How do you fix Claude's too many requests error?

Run these six steps in order. None of them touches your chats or your plan.

1. Wait a minute, then retry once. Anthropic's Claude Code error reference says "Wait briefly and try again". Hammering retry only sends more requests.
2. Update Claude Code with claude update. Version 2.1.199 and later retry these throttles with backoff before you ever see them. The current release is 2.1.316, from 31 October 2026.
3. Run /status and check the active credential. Anthropic warns that "A stray ANTHROPIC_API_KEY in your environment can route requests through a low-tier key instead of your subscription."
4. Close extra sessions and subagents. For 429s on an API key, the docs say to "avoid running many parallel subagents". In one Max subscriber's log, 114 of 122 minutes with a 429 had six or more desktop sessions open.
5. Switch the model with /model. On the API, each model row has its own requests and tokens per minute, so a smaller model can keep a scripted run going.
6. Check status.claude.com, then report it. If nothing is posted and the error lasts for hours, send the exact message and the request ID.

Here's the wording Claude Code prints: "API Error: Server is temporarily limiting requests (not your usage limit)". Anthropic's explanation is one line: "The API applied a short-lived throttle that is unrelated to your plan quota." Claude Code tells it apart from a real limit "by the absence of the unified quota headers a real limit response carries". In the Code tab of the Claude desktop app, a 30 May 2026 bug report shows the same throttle as "Server is temporarily limiting requests" over a line that reads "Too many requests right now". That's the "claude too many requests right now" people search for.

We searched Anthropic's public Claude Code issue tracker on 2 November 2026 for every 2026 issue whose title or body contains "temporarily limiting requests". That returned 410. One account filed 275 of them, and its late July reports all named Claude Code 2.1.114, a build from April. We left that account out. The other 135 came from 125 different people, and 129 of them were opened before July. The changelog entry for 2.1.199 reads: "Transient server rate-limit errors (429s unrelated to your usage limit) are now retried automatically with backoff for subscribers instead of failing the turn". The throttle didn't go away. A September report counted 195 retried 429s in one day. Read the drop as fewer failed turns, not fewer 429s.

Which Claude too many requests message are you seeing?

The wording tells you whose limit you hit. This table covers Claude Code, the Claude desktop app and the Claude API, as read on 2 November 2026.

Anthropic's error message guide for the Claude apps lists no "too many requests" message at all. Its limit-reached messages each name a reset time; only its "approaching" warnings don't. So if your message names a time, you hit your plan, and Claude usage limit reached has the fixes. If it says overloaded or 529, that's capacity, covered in Claude overloaded.

Why does Claude say too many requests when your usage is low?

Claude says too many requests at low usage because the throttle counts how fast requests arrive, not how much of your plan you've spent. Anthropic's rate limits page says limits can bite over short spans: "a rate of 60 requests per minute (RPM) might be enforced as 1 request per second." So a burst of calls can fail while your average for the minute looks fine.

The bug reports show it plainly. A Max subscriber on 30 May 2026 got the throttle on the day their limits had reset. An 18 August report saw persistent 429s at 1% usage. A 24 September report, at 26 to 27% plan usage, hit a 429 in 14% of session-minutes with six or more desktop sessions open. With five or fewer, it was under 1%.

Sudden growth trips it too. Anthropic's API errors page says "if your organization has a sharp increase in usage, you might see 429 errors because of acceleration limits on the API." Its advice is to "increase your traffic gradually and maintain consistent usage patterns." A Claude Code workflow that fans out to many subagents at once sends that kind of spike. Anthropic doesn't publish the numbers behind the throttle for those who are on a claude.ai plan.

How long does Claude's too many requests error last?

Claude's throttle usually clears in seconds to minutes. Claude Code retries transient 429s up to 10 times with exponential backoff, and it shows Retrying in Ns · attempt x/y while it waits. You only see the error once those retries run out.

Some reports ran much longer. The 30 May reporter said it lasted "two days", and that a new session let them work for "3 hours at least" before it came back. A 15 June report said the throttle had held "for the last 24 hours" on Opus 4.8. In the May thread, a new session and a different model got people working again. If it's still there after an hour with one session open, it isn't yours to fix.

For jobs that run while you're away, Claude Code has a switch for this. Set CLAUDE_CODE_RETRY_WATCHDOG=1 and it retries 429 and 529 errors indefinitely. Cap the wait with CLAUDE_CODE_RETRY_WATCHDOG_MAX_WAIT_MS, for example 3600000 for an hour. Set CLAUDE_CODE_MAX_RETRIES=2 instead if a script should fail fast.

Does paying for Claude Max stop too many requests?

No. The throttle is "unrelated to your plan quota", and the reports above came from Pro and Max subscribers. As of November 2026 Claude costs $20 a month on Pro, $100 on Max and $200 on Ultra, per Anthropic's pricing page. Max lists "Priority access at high traffic times". Anthropic doesn't say if that covers the throttle or not.

Whizi, which publishes this page, runs the same kind of guard and prints the numbers: 10 chat messages a minute and 60 an hour per user, the same on every plan. Its per-minute error reads "Too many requests. Please wait and try again." and a refused request spends none of your allowance (Whizi rate limits). Where Whizi loses: 60 an hour is a hard ceiling, and Claude Max prints no hourly number at all.

How do you fix a 429 rate_limit_error from the Claude API?

Read the error body before you retry, because the Claude API sends 429 rate_limit_error for more than one problem. Anthropic's errors page says your organization "has hit a rate limit, reached its usage tier's monthly spend cap, or reached a spend limit on the Claude Code workspace." Only some of those clear with time.

Two details catch people out. First, a spend limit you set yourself isn't a 429. It's a 400 invalid_request_error that starts "You have reached your specified API usage limits". Second, the spend caps are $500 a month on Start, $1,000 on Build and $200,000 on Scale. At the cap, "API usage pauses until 00:00 UTC on the first day of the next month", and even the SDK's automatic retries fail until then.

The limits belong to your organization as a whole, not to one key. On the Start tier, Claude Opus 5.5, Sonnet 5.5 and Haiku 5.5 each get 4,000 requests a minute and 2,000,000 input tokens a minute, as of November 2026. New organizations may start lower, in an Evaluation tier. On most models the tokens you read from the cache don't count toward input tokens per minute, so prompt caching raises the real ceiling you have. The official SDK retries 429s twice by default and honors retry-after. Watch the anthropic-ratelimit-requests-remaining header in each response, and slow down before it gets to zero.