AI NewsModels & agentsAnnouncement
GPT-Live-1.1 reaches the API with a cheaper standby mode
OpenAI put GPT-Live-1.1 in the API. A new standby mode pauses the model during silence and bills $0.02 a minute instead of $0.05.

Why it mattersA voice agent held open through long pauses now costs less than half what it did. The session stays live, so the model answers at once when the user speaks.
OpenAI put its real-time voice model in the API on 20 November 2026. The company says GPT-Live-1.1 has powered voice in ChatGPT Plus and Pro since 15 October. Developers can now call the same model from code.
A new configuration called standby changes the cost shape. A caller sets an idle timeout between 5 and 60 seconds. When the user stops talking, the model pauses and the WebRTC call stays open. When the user speaks again, the model answers at once.
The prices
OpenAI's documentation gives two rates. An active session, where the model is working, costs $0.05 a minute. A standby session, during silence, costs $0.02 a minute. Backend model calls and tool calls are billed on their own. The rates apply to GPT-Live-1 as well.
What the model does better than GPT-Live-1
OpenAI says the new version transcribes speech more accurately and holds a more consistent accent and voice across languages. The company also says the model understands letters and numbers more reliably. That matters for a caller reading out an order number or an email address. These are OpenAI's own figures, with no third-party test attached.
New voices
Seven new voice names are now available: calandria, compas, destello, halo, hornero, luciernaga, and pearl. OpenAI says the full set is 38 voices across the two Live models. The two Spanish-language names and the Portuguese one are the clue to the better-accent claim: the company says French, German, Hindi, Mexican Spanish, Korean and Japanese all improved.
How a team adopts it
Existing GPT-Live-1 code switches by changing the model ID. The Live endpoint and the client delegation pattern are unchanged. Client delegation lets the voice model ask a text model for help mid-call. The voice session keeps running at the same time. OpenAI says that delegated call is billed on its own, through the AI Gateway or the API.
What standby changes in a real build
A support agent that waits for a human to think used to burn the full $0.05 every minute. The new rate cuts that to $0.02 while the caller is silent. Take a one-hour call with four minutes of actual speech. At OpenAI's rates, the bill drops from about $3 to about $1.20. The model is still the full GPT-Live-1.1 when the user speaks. No reconnect, no cold start, no state loss.
OpenAI has not published rate limits for GPT-Live-1.1 in this post. A build that depends on many concurrent sessions should check the account's own limits.
Source
OpenAI Community: GPT-Live-1.1 has arrived in the API and the GPT-Live-1.1 model page.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.


