Claude 529 Overloaded
- Source
- openrouter.ai
- Author
- OpenRouter editorial sitemap
- Date
Context
A 529 error from Claude is a capacity problem on Anthropic's side, not a problem with your account. Anthropic's API error reference describes the overloaded_error as the API being temporarily overloaded and says it can happen when traffic is high across all users. That separates it from a 429 rate_limit_error, which the same page ties to your own organization hitting a rate limit or spend cap. A higher plan can lift a 429; it does nothing for a 529, because the shortage is in total demand for the model.
OpenRouter's guide lays out three responses, to try in order. First, retry with exponential backoff and some randomness, which Anthropic's SDKs already do (twice by default) and Claude Code does, by default up to 10 times, before it shows "Repeated 529 Overloaded errors", per its error reference. That page also says a 529 does not count against your quota and suggests running /model to switch models, since capacity is tracked per model. Retries handle short spikes but only add waiting when an overload lasts minutes.
The guide's main point is the second step: the same Claude model runs in more than one place. OpenRouter lists five providers for Sonnet 5.5 (Anthropic, Amazon Bedrock, Google Vertex AI, Azure and Claude Platform on AWS), with the same $2 per million input on global endpoints in its October 21 check. Its default routing retries a failed request at the next provider, and each response names the provider that served it. A third step, a fallback model such as Haiku 5.5, covers the rare case when every provider is busy, at the price of a smaller model answering.
Read this as a vendor tutorial with a product to sell. OpenRouter's retry demo used a local stand-in server, since a real overload can't be triggered on demand, and its live failover examples all show Anthropic answering, so the guide does not demonstrate an actual failover during an outage. Its Claude Code test was nine small bug-fix sessions that all completed. The underlying advice holds regardless of vendor: a 529 is cured by more capacity, either later or somewhere else, not by changing your key or plan.
- A 529 overloaded_error means Anthropic's API lacks capacity for that model across all users, so it is not about your key or plan and does not count against quota. A higher tier prevents 429 rate-limit errors, which come from your own usage, but not 529s.
- Anthropic's SDKs already retry 5xx errors with exponential backoff, twice by default, and Claude Code retries up to 10 times before showing Repeated 529 Overloaded errors. OpenRouter notes more retries only lengthen the wait when a model stays busy.
- Because Anthropic tracks capacity per model, switching models with /model is the quickest no-setup fix in Claude Code. OpenRouter lists Sonnet, Opus and Haiku 5.5 on five providers each, with Sonnet 5.5 input at $2.00 per million on every global endpoint.
- On OpenRouter, setting a provider order disables default load balancing, while allow_fallbacks false or a restrictive only list stops failover entirely; model fallbacks trigger on any error from the first model, including rate limits and -length errors.
- token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
- context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
Checking sign-in…
Loading comments…