Claude Code 529 Overloaded Error: Retry, Backoff & Status Checks

Paste your error / 粘贴报错原文

Paste an error to match it against the wiki. / 粘贴报错后自动匹配。

Error text / 报错原文

API Error: Repeated 529 Overloaded errors
API Error: 500
Internal server error
Request timed out

What it means

Claude Code reached Anthropic, but capacity or another server-side condition persisted after its automatic retry loop. The current error reference says transient server errors, 529 responses, request timeouts, temporary 429 throttles and dropped connections are retried up to 10 times by default with exponential backoff. If you see Repeated 529, those retries have already been exhausted.

529 is not your usage limit and does not count against quota. A session, weekly or model allowance message belongs to the separate usage-limit guide.

Separate 529, 500, timeout and quota

Visible message	Likely layer	Best first response
`Repeated 529 Overloaded`	Temporary model/service capacity	Check status, wait, or switch model with `/model`.
`500 Internal server error`	Anthropic infrastructure	Pause and retry; use `/feedback` if persistent without an incident.
`Request timed out`	Load, long output, network or proxy	Retry a smaller task and inspect the network before changing timeout.
Session / weekly / usage limit	Account allowance	Run `/usage` and check reset time; this is not 529.

Fastest recovery after retries are exhausted

Check status.claude.com for an active incident or capacity notice.
Wait a few minutes. The original prompt remains in the conversation, so send try again instead of repasting a large message.
Run /model and switch model if offered; capacity is tracked per model.
If the exact error is a timeout, split the task and verify proxy, VPN and streaming stability.

Retry and timeout controls

Variable	Documented default	When to change it
`CLAUDE_CODE_MAX_RETRIES`	10	Lower it for fail-fast scripts. Raise it only when deliberately waiting longer through incidents.
`API_TIMEOUT_MS`	600000 ms	Raise only for a known slow proxy/network or genuinely long response.

Increasing both values blindly can turn a short outage into a much longer silent wait.

What not to do

Do not delete authentication files for a 529 or isolated 5xx response
Do not confuse capacity with a session or weekly allowance limit
Do not hammer manual retries immediately after the built-in exponential backoff failed
Do not assume every timeout is an MCP problem
Do not perform destructive edits while the agent is repeatedly losing the response stream

How to capture a useful report

Record the exact timestamp, model, Claude Code version, status-page state and whether another model works. For a persistent unreported 5xx, use /feedback; it can include request details needed for investigation.

claude --version
/model
/status
Get-ChildItem Env:CLAUDE_CODE_MAX_RETRIES,Env:API_TIMEOUT_MS -ErrorAction SilentlyContinue

Frequently asked questions

Does 529 consume my quota?

No. The official error reference explicitly separates overloaded capacity from account usage limits.

How many retries already happened?

Up to 10 by default. The visible repeated-529 message appears after Claude Code’s automatic exponential-backoff loop.

Should I raise the retry count?

Only when waiting longer is desirable. Automated scripts often benefit from a lower value so another system can handle the failure.

Is a timeout always an outage?

No. Load and large responses can contribute, but a slow proxy, VPN or dropped connection can produce the same symptom.

Related errors

Related troubleshooting resources

Sources

Reviewed against current official documentation on July 31, 2026.