Skip to main content
This page covers the rules shared by every model call and explains which protocol to choose and how to start safely.
Confirm capability firstAvailable models change with your API key group, model restrictions, and live service state. The whole-site Model Marketplace is not a promise that every model is callable by a particular key.

​
Three required values

For a standard OpenAI-style client, use:
For manual HTTP calls, send the key in Authorization:
Never put a complete key in frontend code, screenshots, chat history, or a Git repository. Client URL rules vary: Claude Code uses the site address without /v1, while Codex CLI uses its Provider configuration. Follow the relevant client tool guide.

​
List models visible to this key

Copy a model ID from data[].id. This is the set currently visible to that key, not a full platform catalog. data[].supported_endpoint_types lists the request protocols currently available for that model. Use only a path whose type is listed: openai for Chat Completions, anthropic for Messages, openai-response for Responses, and image-generation for image generation. If more than one type is listed, choose the matching one for your client.

​
Select the right protocol

The same model name does not mean every request format is accepted. Confirm the protocol in the API and model capability matrix, then read:

​
Minimal verification and retries

  1. List models, then send one short text request or minimum image request.
  2. Confirm the model ID, key group, and path before adding optional fields, streaming, or batch work.
  3. Keep the error text, time, and Request ID on failure. Do not replay an unconfirmed request at high frequency.
For 401, check the key first. For 404, check a missing or repeated /v1. For 400, check the model, protocol, and fields. See the self-check guide for more.