Claude Code Repeated 529 Errors: What to Do
When Claude Code shows API Error: Repeated 529 Overloaded errors, the Anthropic API has temporarily run out of capacity. It does not mean your API key or project code is wrong.
Wait before replacing credentials or changing configuration. Capacity can vary by model, so one model may be unavailable while another still responds.
429, 500, and 529 are different errors
| Code | Name | Meaning |
|---|---|---|
| 429 | rate_limit_error | Your API key or project is sending requests too quickly; read the accompanying message |
| 500 | api_error | An internal Anthropic system error |
| 529 | overloaded_error | Temporary API overload |
A 429 does not automatically mean you exhausted a paid allowance. You've hit your session/weekly limit refers to a subscription session or weekly limit. Request rejected (429) points to an API key or project rate limit. Server is temporarily limiting requests (not your usage limit) means the service is throttling traffic rather than enforcing your account quota.
For both 500 and 529 errors, the standard response is to wait and retry.
What Repeated 529 tells you
Claude Code automatically retries overloads, server errors, timeouts, and temporary 429 responses with increasing delays. The default is up to 10 attempts with exponential backoff, although an environment variable can change that count. If Repeated 529 remains on screen, those retries have already been exhausted. Wait a few minutes, then resend the last prompt.
If the error continues, check the Claude status page. An incident such as “Elevated errors” means the fix must come from the service side. If Claude Code connects through Amazon Bedrock, Google Cloud, Microsoft Foundry, or a custom gateway, also check the status of the provider named in the error.
Continue with another model
If the status page reports no broad outage but one model keeps returning 529, enter /model and try another option. For example, test whether Sonnet or Haiku responds when Opus does not.
If the status page is clear and the error persists, run /feedback inside Claude Code so the report includes the relevant request details. For a direct API call, include the response request_id in your support request. Until then, leave the configuration alone and retry at wider intervals.