Live updates
OpenAI opens GPT-Live-1.1 to developers, with idle time billed at 2 cents a minute
First seen on X 33 hours ago@OpenAIDevs ♥ 1,20033 views
OpenAI has made GPT-Live-1.1, the voice model behind voice conversations in ChatGPT Plus and Pro since Oct 15, available in its API. The full-duplex model can listen while it speaks, and OpenAI says it transcribes speech more accurately than GPT-Live-1 and keeps a more consistent accent and voice across languages.
Sessions are billed by duration, according to OpenAI's developer account and API docs on Nov 19: $0.05 a minute while active, or $3 an hour. A new standby setting pauses inference after 5 to 60 seconds of silence and drops the rate to $0.02 a minute, then resumes when someone speaks without breaking the conversation. For voice agents with long silences, such as support lines, that cuts the cost of idle stretches by 60%. Standby works with both GPT-Live-1 and GPT-Live-1.1 on WebRTC sessions. OpenAI says the model is better at catching details like letters and numbers and can be told in plain language when to hand tasks to a backend model; backend model and tool usage is billed separately. Early reactions were mixed. One German developer replied that it "doesn't sound too good in german" and that they hadn't noticed any improvement since using it in ChatGPT.
Comments 0
Select a sentence in the story (long-press on phones) to comment on it.
“GPT-Live-1.1 is now available in the API.”
