Issue
I’m trying to use a Prompt Template (Prompt ID) with the Realtime API via session.prompt, but the API always returns a strict model mismatch error:
Session model 'gpt-realtime' does not match prompt model 'gpt-4.1'.
Switching to realtime preview models (e.g., gpt-4.1-realtime-preview-*) produces the same error:
Session model 'gpt-4.1-realtime-preview-2024-12-17' does not match prompt model 'gpt-4.1'.
Cause
The Prompt Template UI does not list any realtime-capable models.
You can only create prompts for text models like gpt-4.1.
As a result, no prompt template can match any realtime model, and Prompt IDs cannot be used with Realtime sessions.
Expected
Since session.prompt is documented for the Realtime API, I expected one of:
- Prompt templates supporting realtime models
- Or realtime models being selectable in the prompt editor
- Or realtime models accepting text prompt templates
Actual
- Prompt templates only support text models
- Realtime API only accepts realtime models
- No matching model exists → PromptID is unusable with Realtime API
Question
Is PromptID officially unsupported for the Realtime API right now?
If yes, is support planned, or should we avoid PromptID for realtime use cases?
curl -s -X POST "https://api.openai.com/v1/realtime/client_secrets" ^
-H "Authorization: Bearer <API KEY>" ^
-H "Content-Type: application/json" ^
-d "{\"session\":{\"type\":\"realtime\",\"model\":\"gpt-4.1-realtime-preview-2024-12-17\",\"prompt\":{\"id\":\"<PROMPT_ID>\"}}}"
{
"error": {
"message": "Session model 'gpt-realtime' does not match prompt model 'gpt-4.1'. ",
"type": "invalid_request_error",
"param": null,
"code": "invalid_parameter"
}
}
