OpenAI has put GPT-Live-1.1 in the API under the model ID gpt-live-1.1, MarkTechPost reports. It is a full-duplex voice model for multilingual conversations, meaning it can listen while it speaks. OpenAI says 1.1 has more natural and consistent accents, more accurate transcription, better handling of precise details such as letters and numbers, and more reliable task completion than GPT-Live-1. The company has not published benchmark scores for those claims, so I am treating them as self-reported. Inputs and outputs are audio and text. Image, video, and structured outputs are not supported.
The other headline is standby, a cheaper billing state for idle WebRTC sessions. Inference pauses during silence while the connection stays open, and OpenAI’s pricing, as reported, is $0.05 per minute while active and $0.02 per minute in standby. The idle timeout is configurable between 5 and 60 seconds. Standby is documented for both GPT-Live-1 and 1.1. Backend reasoning and tool calls, which can be delegated to a model such as GPT-6 Astra or Luna, are billed separately. I have not used the API for this post.
My take
Cheaper idle time is the practical change if you already run voice sessions that spend a lot of the call waiting. The quality claims will have to be tested, because OpenAI did not publish numbers this time. If you try 1.1, read the model page and measure a real call, including whatever the backend model costs.
God Bless, and Use the Source!