Anthropic API 529 Error: How to Fix It

HTTP 529 with overloaded_error means Anthropic's capacity for the requested model is saturated. It is not caused by your API key, your account, or your payload.

Retry with exponential backoff and jitter

Retry immediately and you make the problem worse. Use delays of roughly 1s, 2s, 4s, 8s with random jitter, capped at five attempts.

for attempt in range(5):
    r = call_claude()
    if r.status_code != 529: break
    sleep((2 ** attempt) + random.random())

Fall back to another model

Capacity pressure is usually model-specific. Configure a fallback to a smaller or older Claude model so user-facing requests still complete.

Shorten and batch your requests

Very large context windows consume more capacity per request and are more likely to be shed. Trim system prompts and cache repeated context.

Queue non-urgent work

Move batch jobs off peak hours and process them through a queue with a concurrency limit, so a burst of 529s does not cascade.

Distinguish 529 from 429

A 429 is your own rate limit and needs different handling — see Claude API rate limit exceeded. If requests fail before reaching the model at all, check API key authentication failing.

Full checklist: Anthropic API 529 overloaded fix page.