Is this error coming from 9Coding?

1 · Check the address and the request id#

9Coding exposes two APIs, and their error formats differ. Only a request sent to one of these two addresses can get an error returned by 9Coding.

Inference gateway · api.9coding.com: serves every model call, including OpenAI-compatible endpoints such as /v1/chat/completions and the /v1/messages endpoint that Claude Code uses. A missing or invalid key returns:

{
  "error": {
    "code": "",
    "message": "Invalid token (request id: 2026…)",
    "type": "new_api_error"
  }
}

How to recognise it: the request went to api.9coding.com and message ends with (request id: …); for a missing or invalid key, type is new_api_error. The request id starts with a UTC timestamp (such as 2026…; Beijing time is 8 hours ahead), and the gateway's x-oneapi-request-id response header carries the same id, visible with curl -i. Some errors, such as upstream errors that 9Coding forwards, may not carry a request id in message. Other relays running the same open-source gateway return the same format, so the address is what decides.

Console API · 9coding.com/api: serves the website and console, including sign-in, key management, top-ups and usage. Model calls never go through it. A request without a valid sign-in returns:

{
  "message": "missing bearer token",
  "success": false
}

How to recognise it: message and success side by side, with success set to false. These errors appear in requests made by the website and console (for example, in the Network panel of the browser's developer tools) and are unrelated to model calls; when the sign-in has expired, signing in to the console again resolves it.

2 · Compare with known third-party errors#

Only formats distinctive enough to identify the source with confidence are listed.

• Browserbase: Missing x-bb-api-key header
  Signs: statusCode, error and message sit side by side; x-bb-api-key is the auth header of Browserbase, a cloud browser service.
  Fix: configure Browserbase's own key — see below.
• OpenAI API: Incorrect API key provided
  Signs: points to platform.openai.com; code is invalid_api_key.
  Fix: the request went straight to the official OpenAI API; set the base URL to https://api.9coding.com/v1.
• Anthropic API: "request_id":"req_…"
  Signs: a top-level request_id field starting with req_; no (request id: …) at the end of message.
  Fix: the request went straight to the official Anthropic API and bypassed 9Coding; check that ANTHROPIC_BASE_URL is in effect.
• Google API: API key not valid
  Signs: an error from an official Google API (such as the Gemini API) with "status":"INVALID_ARGUMENT" and a reason of API_KEY_INVALID in details.
  Fix: the request went straight to Google, so find the tool that sent it. For Gemini CLI, set GOOGLE_GEMINI_BASE_URL as in the Gemini guide; other tools that call Google directly (such as the Browserbase MCP server — see below) need a Google key.
• The client: Could not resolve authentication method
  Signs: the client failed before sending the request; there is no HTTP response.
  Fix: set ANTHROPIC_AUTH_TOKEN — see 401 Unauthorized.
• Local machine or network: ECONNREFUSED · ENOTFOUND · ETIMEDOUT
  Signs: no status code and no response body.
  Fix: see Connection error.

If the key went to a third party: in the OpenAI, Anthropic and Google cases, if the request carried a 9Coding key (for example, the key echoed in OpenAI's error starts with sk-9c-), that key has been sent to the other service. After fixing the address, delete the key in the 9Coding console and create a new one.

3 · Confirm the 9Coding side with one curl request#

The example uses the Claude Code variable. Use the key that the failing tool is actually configured with; for other clients, replace the variable with the key entered in that tool.

curl -s https://api.9coding.com/v1/models \
  -H "Authorization: Bearer $ANTHROPIC_AUTH_TOKEN"

curl.exe -s https://api.9coding.com/v1/models `
  -H "Authorization: Bearer $env:ANTHROPIC_AUTH_TOKEN"

• A model list comes back → this key and address are valid. If the original error carries (request id: …), or is a model-not-found, rate-limit, balance or context-length error, it still came from 9Coding; look up the matching page in the troubleshooting index. If the original error is an authentication error without 9Coding's markers, it came from another service in the toolchain; use step 2 to find which.
• An error with (request id: …) comes back → it came from 9Coding. Invalid token means the key is invalid or no key was sent (an empty variable gives the same result) — see 401 Unauthorized; for anything else, look up the matching page in the troubleshooting index.
• No output at all → curl got no response; this is a network problem — see Connection error.