TL;DR

A 529 Overloaded error means Anthropic's API is temporarily at capacity across all users, and the docs say capacity is tracked per model. It does not count against your quota. Claude Code has already retried by the time you see it. Check status.claude.com, wait a few minutes, or run /model to switch. A 429 is a different error.

A 529 means Anthropic's API is temporarily at capacity across all users, and the docs say capacity is tracked per model. It is not your usage limit. Claude Code has already retried several times before it shows you the message, so the moves the docs list are to check status.claude.com, wait a few minutes, or switch model with /model. Last verified 14 November 2026 against the official Claude Code error reference, with the npm registry listing 2.1.325 as the latest Claude Code version.

This guide covers the exact messages, what each one means, and the retry settings the docs list. For a live view of incidents, the Claude Code status hub tracks the same page.

What does a 529 error mean in Claude Code?

The Claude Code error reference lists this message (the docs print a dash after "capacity"; it is shown here with a full stop):

API Error: Repeated 529 Overloaded errors. The API is at capacity. This is usually temporary. Try again in a moment. If it persists, check https://status.claude.com.

The docs explain that the API is temporarily at capacity across all users, and that "a 529 is not your usage limit and doesn't count against your quota." You will also see model-specific wording such as Opus is experiencing high load, please use /model to switch to Sonnet.

How to tell an outage from a local problem

Open status.claude.com and look at the Claude Code component and the open incidents. An incident that names a higher error rate or overloaded errors on your model matches a 529. The docs do not say how quickly a capacity problem appears on the status page, so a clean page does not prove the API is healthy.

A local fault looks different. A request that never reaches the API fails with a network message, and the docs describe a banner that reads Waiting for API response · will retry in … · check your network after 20 seconds without receiving any data (the docs say it clears on its own once data resumes or a retry succeeds). A 529 is an overloaded response from the API, and the docs describe it as capacity across all users rather than a problem on your side.

What should you do when you get a 529?

  1. Wait a few minutes and send the prompt again. The docs say to try again in a few minutes.
  2. Run /model and pick a different model. The docs state that "capacity is tracked per model", so a busy model does not mean every model is busy.
  3. Check the status page, or the provider status page named in the message if you use a cloud provider.
  4. If you run unattended sessions, read the retry settings below before changing anything.

The error reference does not list restarting Claude Code, reinstalling it or logging out as a fix for a 529.

How does Claude Code retry before showing the error?

The docs say Claude Code retries transient failures up to 10 times with exponential backoff before it shows an error, and the spinner reads Retrying in Ns · attempt x/y. These variables tune that behaviour:

  • CLAUDE_CODE_OVERLOADED_RETRY_BASE_DELAY_MS: default 500, can be raised to 32000. Needs v2.1.292 or later.
  • CLAUDE_CODE_OVERLOADED_RETRY_MAX_DELAY_MS: default 32000, can be raised to 300000. Values above 64000 need v2.1.323 or later, and the variable itself needs v2.1.296 or later.
  • CLAUDE_CODE_RETRY_WATCHDOG=1: retries 429 and 529 capacity errors indefinitely instead of failing after CLAUDE_CODE_MAX_RETRIES attempts. The docs describe it for unattended sessions such as CI. With it set to 1, the two delay variables above have no effect.

As a cautious approach, set them in your shell for one run first, for example export CLAUDE_CODE_OVERLOADED_RETRY_MAX_DELAY_MS=120000 on macOS or Linux. These settings come from the docs and were not run against a real 529, because that cannot be triggered on demand.

How is a 529 different from a 429?

They sound alike and mean different things. A 529 is capacity on Anthropic's side. A 429 is a limit applied to you: a plan allowance, an API key limit or a short throttle. The table sets the messages side by side, with the docs as the source for each row.

What you seeLikely causeWhat to do
Repeated 529 Overloaded errorsAPI at capacity for all usersCheck status, wait, or switch with /model
Opus is experiencing high loadOne model is busyRun /model and choose another model
Server is temporarily limiting requests (not your usage limit)Short throttle, unrelated to your quotaWait briefly and retry
Request rejected (429)Rate limit on your API key or cloud projectRun /status and check the active credential and limits
You've hit your session limitPlan usage allowance used upWait for the reset time shown, or run /usage
Waiting for API response · check your networkNo data received for 20 seconds or more; not a failure yetWait for it to clear; if it reappears on every attempt, the docs say to treat it as a network issue

The usage-limit side has its own guide: Claude Code 429 and usage limit reached.

What if the 529 keeps coming back?

Because the docs say capacity is tracked per model, switching with /model is the documented answer when one model keeps failing. The docs describe /feedback as the way to report a bug with session context attached. In the error reference it is recommended for a persistent 500 with no posted incident. It does not say that for 529s, so treat it as a general reporting route.

Two public GitHub reports show the message in the wild. One user reported failures on Opus over the previous hour on Claude Code 2.0.69 in issue 13990, and another reported it on 2.1.145 in issue 63051. The first was closed by its reporter within the hour, and the second was closed by a bot as a duplicate of another issue. Neither has a maintainer reply. They show the message was reported, nothing more, and neither is documented behaviour.

If the failure is a login message rather than a capacity one, see what the authentication service error means.

More practical writing for people building with AI tools is on the Lead Source blog.

Fixed? Two optional things

Check Anthropic’s status and get an alert when an incident is resolved: Claude Code status and alerts, an independent resource by Lead Source.

Official source: status.claude.com.

Written by the team at Lead Source, where the day job is lead source tracking: working out which marketing actually brings customers.