OpenBridgerGuides
中文 Open console

Error codes

When the gateway itself rejects a call, the body looks like this:

{
  "error": {
    "message": "explanation (request id: REQUEST_ID)",
    "type": "new_api_error",
    "code": "error code, or an empty string"
  }
}

The X-Oneapi-Request-Id response header and the request id in the body are the same value. Send that id to support. Do not send the full API key.

An upstream error can keep the upstream type and HTTP status. It is not always new_api_error. This table is only what callers hit most often from the gateway itself.

Authentication and permission

HTTP code Meaning What to do
401 empty Invalid token The key is wrong, disabled, expired, or that key's own credit is used up. A funded wallet does not change this. Copy the sk- key again, or raise that key's credit.
401 empty Token was not provided There is no Authorization header. It should be Bearer plus the key.
403 empty No access to this group, or the group is retired Change the key group back to the account group or auto.
403 access_denied The IP is not on the key allowlist, or the client IP cannot be parsed Edit or clear the allowlist.
403 empty The account is banned Email support from the registration address. Do not keep retrying the same key.
403 empty A normal user cannot pin a channel Do not append -channelId to the key.

An Anthropic client may send the key in x-api-key. The gateway accepts that only on /v1/messages and /v1/models. Other paths need Authorization: Bearer.

Credit and billing

HTTP code Message What to do
403 insufficient_user_quota User quota is insufficient, or the pre-charge failed The wallet cannot cover this reserve. Top up, redeem a code, or send a shorter input.
403 insufficient_user_quota Subscription credit is insufficient or no subscription is configured The preference is Subscription Only, or the subscription balance is too low. Switch to Wallet First or add wallet credit. See Subscriptions.

Insufficient credit is 403, not 429.

Model and request

HTTP code Message What to do
400 empty or a request error The model name is empty, or the body is not valid JSON Check the JSON and the model field.
503 model_not_found No channel is available for this group and model Copy the ID from Model Square again and check the key's group. This is not a wrong-URL 404.
501 api_not_implemented API not implemented That path is not implemented. Image variations, files, and fine-tunes are in this set.
429 empty You reached the request limit, including failed tries when the message says so This appears only when the application rate limit is enabled. The window and the count are in the message. Lower concurrency and wait.
500 empty An internal failure such as a database error Keep the request id and contact support. Do not retry the same key in a tight loop.

An upstream limit can be the upstream's own 429 and body. That is not the local rate-limit rule above. Retry later, or pick another model that Model Square still lists.

Looks like a gateway error, and is not

What you see Actual cause
The path is /v1/v1/chat/completions The Base URL already contains /v1, and the client added it again. OpenAI-compatible clients use https://openbridger.com/v1.
Claude Code calls the wrong host ANTHROPIC_BASE_URL must be https://openbridger.com, without /v1. The client requests /v1/messages itself.
The model list is empty, but curl /v1/models returns data The client filtered unknown IDs. Paste the full ID from Model Square.
The browser reports CORS, or the key leaks Do not put the key in a web page. Call from your own server.

Related

Type to search