Errors¶
Errors come in the format of the API you called. The OpenAI APIs carry a
code, listed below. The Anthropic Messages API has no field for a code, so
its errors carry a type that follows the status, such as billing_error for
402, and the message says the rest.
| Status | Code | What it means |
|---|---|---|
| 400 | context_length_exceeded |
The conversation is too long for the model. Agents shorten it and retry. |
| 400 | missing_model, invalid_body, unsupported_parameter |
The request is not valid. |
| 401 | missing_api_key, invalid_api_key |
No key, or a key the gateway does not know. |
| 401 | expired_api_key, revoked_api_key |
The key has expired or was revoked. Get a new one. |
| 402 | budget_exceeded |
A budget is used up. The message says when it resets. |
| 402 | credit_exhausted |
The prepaid credit is used up. See Billing. |
| 403 | filter_refused |
A filter stopped the request. The message gives its reason. |
| 403 | org_limited |
The organisation must pay once before it can use this. |
| 403 | subscriptions_off |
The gateway does not take Claude plans. |
| 404 | model_not_found |
No such model, or the key may not use it. |
| 404 | wrong_base_url |
The client's base URL is missing /api. |
| 413 | request_too_large |
The request is larger than the gateway accepts. |
| 413 | filter_input_too_large |
The request is too long for a filter's model. |
| 429 | rate_limit_exceeded, token_rate_limit_exceeded |
Too many requests or tokens this minute. Wait for Retry-After. |
| 429 | requests_quota_exceeded, tokens_quota_exceeded |
The daily or monthly limit is used up. |
| 502 | upstream_unavailable, upstream_error |
The model could not be reached, or failed. |
| 502 | filter_failed |
A filter's model failed, so the request was stopped. |
| 503 | no_backend, filter_unavailable |
A model or filter cannot run right now. |
| 503 | router_undecided, router_destination_unavailable |
A router could not place the request. |
| 503 | control_plane_unavailable |
The gateway lost its database for longer than its cache lasts. |
Every error answer carries an X-Request-Id. Search for it on the
Requests screen to see the full story.