Troubleshooting · Claude Code

Claude Code API Error 529 Overloaded — What It Means

Claude Code prints API Error: 529 Overloaded when upstream capacity for the selected GPT or Gemini model is saturated. Why it is not your request's fault, why it clusters, and what actually helps.

What you see

API Error: 529 {"type":"error","error":{"type":"overloaded_error","message":"Overloaded"}}

Why it happens

  • Upstream capacity for the selected GPT or Gemini model is temporarily saturated. The endpoint reports this as 529 overloaded_error, the status Claude Code already retries. It describes the service at that moment, not your request — nothing in your payload caused it.
  • It clusters during incidents and peak hours: the same request typically succeeds minutes later with no change on your side.

How to fix it

  • Let Claude Code retry — it backs off automatically. If a run keeps dying, wait a few minutes rather than hammering the same minute.
  • For long unattended runs, choose a lighter model for mechanical steps: it is generally less contended, so the run survives capacity dips.
  • If 529s persist for a long stretch, point ANTHROPIC_MODEL at another GPT or Gemini model — your configuration is almost never the cause.

Related searches

  • claude code overloaded error
  • claude code 529 keeps happening

FAQ

Is a 529 different from a 429?+
Yes. A 429 is your own throughput ceiling; a 529 is upstream capacity. Backoff helps both, but only a 429 responds to reducing your own usage.
Did my prompt cause the 529?+
No. Overloaded is a service-side condition. The identical request usually succeeds once capacity recovers.
Should I switch models when I see 529s?+
For latency-sensitive work, temporarily using another GPT or Gemini model helps because capacity is per model. For everything else, waiting is enough.

Skip the broken setup

apiToken.sale serves GPT and Gemini models through OpenAI-compatible, Messages (Anthropic format) and native Gemini endpoints — standard SDKs, one key, one prepaid balance. Point your tool's base URL at it, and the config on this page works as written.