Claude Code 529 Overloaded, 429 and 500 errors

API Error: Repeated 529 Overloaded errors. The API is at capacity — this is usually temporary. Try again in a moment. If it persists, check your inference gateway (your-provider.example).

529 means the server is overloaded and did not handle the request. Claude Code retries on its own: in our test, claude -p stopped after three 529s in a row, about five seconds, and showed the message above. With retries turned off, the message is API Error: 529 Overloaded. This is a server-side issue, usually temporary…

• On the Anthropic API, it is usually load on Anthropic's side. Check status.claude.com for an incident and try again in a little while.
• On a custom base URL, the address in brackets at the end is the one Claude Code connected to, and that address returned the 529. Either the provider or its upstream is full. Wait, or ask the provider.
• Do not resend right away in a loop. Wait a minute or two.

API Error: Request rejected (429) · the reason the server gave

429 is a rate limit: too many requests in a short time, or too many at once. The text after the dot comes from the server and differs between services. Claude Code retries first: in our test it retried 10 times and showed the error after about three minutes, so it looks stuck in the meantime.

• Several Claude Code sessions at once, or many subagents working in parallel, hit concurrency limits fastest. Run fewer at once, or split the work and run it in sequence.
• Third-party services usually limit concurrency or rate per key. Check your provider's limits; if a team shares one key, give each person their own.
• On the Anthropic API, limits depend on your organization and usage tier, which you can see in the Console.

API Error: 500 Internal server error. This is a server-side issue, usually temporary — try again in a moment. If it persists, check your inference gateway (your-provider.example).

500 is an error inside the server and has nothing to do with your settings. Claude Code retries this 10 times as well, so in our test the message appeared after about three minutes. A single one can be ignored. If it keeps happening, check the service status or contact your provider.

Prompt is too long · the request is ~250000 tokens (limit 200000) …

This is not a server fault: the request is longer than the model's context window. When a conversation has grown long, run /compact to shorten it or /clear to start over. If even the first message is too long, the attached files are too large or too many tools are connected.

429 and 500 are retried 10 times before Claude Code shows them. To see the error at once, set the environment variable CLAUDE_CODE_MAX_RETRIES; 0 turns retries off:

CLAUDE_CODE_MAX_RETRIES=0 claude

Do not leave it at 0 for normal work. Retries are what carry you through short network drops and brief overloads.

401, 403, 404 and connection errors are covered in Claude Code API errors. The base URL and key setup is in Connect Claude Code to your API key.