Home → Help
529 overloaded_error from Claude — what it is and what to do
529 is Anthropic saying the model is over capacity at that instant. Your key, your balance and your rate limit are all irrelevant to it. The request never ran, so you are not billed for it.
What you are seeing
529 {"type":"error","error":{"type":"overloaded_error"}}anthropic.InternalServerError: Error code: 529Bursts of failures that clear on their own within minutesRetrying immediately produces another 529
Why it happens
It is a server-side condition tied to demand on the model, usually concentrated on the newest and largest ones during peak hours. Nothing in your request causes it.
It is often confused with 429. They are opposite problems: 429 means you sent too much, 529 means the provider has too much from everyone. Backing off your own traffic will not clear a 529 any faster, but retrying instantly makes you part of the stampede.
Through a gateway the picture is the same, with one addition: if the gateway holds several upstream accounts, a 529 on one of them may be retried elsewhere before you ever see it.
How to fix it
- Retry with exponential backoff and jitterWait roughly 1, 2, 4, 8 seconds with a random offset. The jitter matters — synchronised retries from every client at once is what keeps the overload alive.
- Have a fallback model for work that can take oneBatch and background jobs can drop to a smaller model on 529 rather than fail. Interactive requests are usually better off waiting.
- Do not treat it as a billing problemTopping up, rotating keys or raising limits changes nothing here. Check the provider's status page if it persists beyond a few minutes.
A 529 that reaches you through APICLAN came from the upstream provider, and it is not billed — an unrun request costs nothing. Requests that do run appear in your usage log with their exact charge.
Related
Last checked 2026-10-01. Written from problems diagnosed on a live
OpenAI-compatible gateway, not collected from other sites.