IndexGPT Realtime
GPT Realtime
GPT Realtime is the API model for live speech in and speech out. You open a realtime session from the docs. It is not a voice mode toggle you buy on top of Plus.
Updated 2026-09-27
The model page is GPT Realtime. Voice chat inside the ChatGPT app is a product surface on your plan. The API model is for an app you build, and the minutes land on API billing.
Before you wire it up
- Use a current model id from that docs page. Names change
- Keep the key on a server. A browser should receive a short-lived client secret, not the account key
- Budget both directions of audio. A quiet session still holds the connection
- Count related text with the token counter only as a sketch. Audio pricing is on the rate card, not in that box
Questions
Is gpt-realtime the same as ChatGPT voice?
Voice in the app uses OpenAI’s speech stack on your plan. gpt-realtime is the API model you call from your own product.
Does Plus include realtime API minutes?
No. Plus is the ChatGPT subscription. Realtime API usage is metered.
Related notes
- OpenAI API and API keys
Create an OpenAI API key at platform.openai.com. Usage is billed apart from ChatGPT Plus. new-api is a separate open-source gateway, not the official API.
- Token counter
Estimate an OpenAI token count at about four English characters per token, then confirm on platform.openai.com/tokenizer before you budget a prompt.
- OpenAI chatbot
An OpenAI chatbot is a chat completion loop plus your instructions. Start on the API, keep the key on a server, and add tools only when a reply is not enough.