Errors

LLMRPM uses structured JSON for gateway-generated failures, while model-originated errors are passed through with their original status and payload.

Error shape

Gateway errors generated by LLMRPM — authentication, subscription, quota, and gateway availability failures — use the OpenAI-style shape below, with matching message, type, and code fields:

Gateway error response

{
  "error": {
    "message": "7-day request limit exceeded. Try again after 2026-09-02T09:14:03.000Z.",
    "type": "quota_exhausted",
    "code": "quota_exhausted"
  }
}

Errors returned by the target model itself are passed through unchanged. Use the HTTP status first, then inspect the provider-specific payload when the error did not originate from the LLMRPM gateway.

Status codes

Status	Meaning	How to handle it
400	The target model rejected your request body — often a missing required field or an invalid parameter for that model.	Check the model-specific error message and fix the request; retrying without changes will fail again.
401	Missing, invalid, or revoked API key.	Check the `Authorization` header and confirm the key is still active in the dashboard.
402	Generation request made without an active subscription.	Subscribe on the pricing page; do not retry until a plan is active.
403	Email not verified, or the usage endpoint was called without an active subscription.	Verify the account email or activate a subscription, depending on the response message.
429	A 5-hour, 24-hour, or 7-day request window is exhausted.	Read the reset time in the message or check `GET /api/v1/usage`, then retry after that boundary.
5xx	Gateway misconfiguration or an unreachable/failing upstream model.	Safe to retry with backoff. Persistent failures should be reported via contact.

Retry guidance

Only 429 and 5xx responses are safe to retry automatically. A quota 429 includes the exact reset time; wait until that boundary rather than repeatedly retrying.

Back off without retry loops

For transient 5xx failures, use exponential backoff and jitter: wait roughly 1s, then 2s, then 4s, capping around 30–60s between attempts. For a quota429, pause until the exhausted window resets.

Do not retry 400, 401, 402, or 403 without changing something first — the request will fail identically every time until the underlying issue (a bad parameter, a bad key, a missing subscription, an unverified email) is fixed.

PreviousUsage & limits Next Integrations