I am using the OpenAI Realtime API. The main workflow (chit-chat and function calls) uses the gpt-4o-realtime-preview model. However, I also generate some out-of-band responses. Before I submit response.create for those, I update the session to use the gpt-4o-mini-realtime-preview model and generate them accordingly. After that, I switch back to the regular model.
My issue is that billing does not show any costs for the gpt-4o-mini-realtime-preview model; it only displays costs for the gpt-4o-realtime-preview model. The tracing is showing that the generated output audio is using gpt-4o-mini-realtime-preview.
1 Like