Skip to main content
This page covers OpenAI Chat Completions only. Before calling, read the API and model capability matrix to confirm that the selected model and protocol match.
Use different sources for different questionsUse the Model Marketplace and the capability matrix for current model availability, token-group access, and protocol selection.

​
Endpoint

​
Minimal request

​
Required headers

​
Minimum fields and streaming

For a normal response, text is usually read from choices[0].message.content; for a streamed response, read each choices[0].delta.content. Verify the model and key with the non-streaming minimum request before enabling stream or adding optional fields. Optional fields such as temperature, tool calls, vision input, and structured output are not guaranteed for every model merely because an API is “OpenAI-compatible.” Use the API and model capability matrix plus Model Marketplace for current model availability and supported capabilities.

​
Common failures

If you see status_code=500, not implemented, do not retry it as a normal 500 at high frequency. It usually means that the model and request protocol do not match. Check whether supported_endpoint_types includes openai, then use the matching guide. For any other error, keep the Request ID and use the self-check guide. Avoid rapid retries.