Market · AI in practice
GPT-Live-1.1 arrives in the OpenAI API with a standby mode at US$ 0.02 a minute
GPT-Live-1.1, a voice model that listens while it speaks, is now in the OpenAI API. With standby, after 5 to 60 seconds without speech inference pauses, the WebRTC call stays open and the per-minute rate drops from US$ 0.05 to US$ 0.02.
What the sources report
GPT-Live-1.1 is OpenAI's new full-duplex voice model, which listens and speaks at the same time. Compared with GPT-Live-1, the company says it transcribes speech more accurately and keeps a more consistent accent and voice across languages. Developers describe in plain language when the model should hand a task to their backend for reasoning or tool use. The model runs in the API's Live sessions and takes audio and text as input.
Standby works for GPT-Live-1 and GPT-Live-1.1 WebRTC sessions. Developers set a timeout of 5 to 60 seconds without an input or output transcript, after which inference pauses. The connection and session ID stay the same, and the conversation resumes with its history when someone speaks again. OpenAI notes that the first response after standby may take a little longer.
Sessions are billed by duration, at US$ 0.05 per minute when active and US$ 0.02 per minute on standby, 60% less. Backend model and tool usage is billed separately. Among the voices listed in OpenAI's guide, two are in Portuguese with a Brazilian accent, Bossa and Tempo.
My take
Customer service calls have a lot of silence in them, with the customer looking for a card, waiting for a code or reading out a number. With idle time cheaper, keeping a voice agent on the line for the whole conversation becomes easier to justify in financial services, and voice is likely to be the first agent interface many bank customers actually use.