Claude "API Error 529 Overloaded": What It Means and What to Do

Why it happens

Anthropic's API documentation defines 529 as overloaded_error: the API is temporarily overloaded, usually because of high traffic across all users. Nothing in your prompt, settings or account causes it.

1. Peak demand on one model. Capacity is tracked per model, so Opus can be overloaded while Sonnet is fine. Claude Code may say so directly, for example: Opus is experiencing high load, please use /model to switch to Sonnet.
2. An active incident. Waves of 529s have been reported many times on the Claude Code tracker, for example the July 2025 outage in issue #3503, where the CLI showed API Error (529 ...) with a Retrying in 1 seconds… (attempt 1/10) countdown.
3. The retry budget ran out. Claude Code retries transient failures up to 10 times with exponential backoff and only then shows Repeated 529 Overloaded errors. Issue #81330 (open) reports that the budget gives up within a few minutes, so a longer overload looks permanent.
4. Compaction hit the overload too. /compact and auto-compact make their own API call, so they can fail with the same 529 while the API is at capacity.
5. A third-party provider is at capacity. On Amazon Bedrock, Google Cloud or a custom gateway, the message names that provider's status page instead of status.claude.com.

The fix

1. 1 Open status.claude.com (or the provider status page named in the message) and check for an active capacity notice or incident.
2. 2 Wait a few minutes, then resend. Your original message is still in the conversation, so for a long prompt you can type: try again.
3. 3 Run /model and pick a different model to keep working while the busy one recovers.
4. 4 Set a fallback chain so overloads switch models automatically: start with claude --fallback-model sonnet,haiku, or add "fallbackModel": ["claude-sonnet-5", "claude-haiku-4-5"] to settings.json. The switch lasts for the current turn only.
5. 5 For unattended runs such as CI, set CLAUDE_CODE_RETRY_WATCHDOG=1 so Claude Code keeps retrying 429 and 529 capacity errors instead of failing. On v2.1.292 or later you can instead raise CLAUDE_CODE_OVERLOADED_RETRY_BASE_DELAY_MS (default 500, max 32000) to spread retries over a longer window.
6. 6 If 529s continue with no posted incident, run /feedback so Anthropic can look at your request details.

529 vs 429 vs 500

A 529 means the whole API (or one model) is at capacity for everyone. A 429 is a rate limit tied to your account or plan, and it does count against you. A 500 is an unexpected failure inside the API. All three are worth a status page check, but only the 429 is fixed by waiting for your own limit window or changing plan.

If you call the API directly, the official SDKs already retry 5xx errors with exponential backoff, twice by default, and accept a max_retries option if you want more attempts during busy periods.

Still failing?

• Check whether every model fails or only one, because a single overloaded model is fixed by /model while a full outage is not.
• If you route through Bedrock, Google Cloud or a gateway, check that provider's status page rather than status.claude.com.
• If /compact itself fails with 529, wait for capacity to return before compacting, since a fallback model with a smaller context window will not be used for compaction.

Related errors

• Claude "Spanner temporarily unavailable"
• Hermes disables a valid key on "overloaded" responses