OpenAI's GPT-Live-1.1 adds a standby mode for idle voice sessions
The updated realtime voice model improves accent consistency and transcription, and a new standby mode cuts the per-minute rate whenever a caller goes quiet.
OpenAI shipped GPT-Live-1.1 in the API yesterday. The update improves pronunciation, accent consistency and transcription in multilingual voice conversations, and OpenAI’s developer team says it also follows instructions and completes tasks more reliably than the original GPT-Live.
The more immediate change for anyone running a voice app is standby mode, now available on WebRTC sessions for both GPT-Live-1 and GPT-Live-1.1. It pauses inference during inactivity, with an idle timeout configurable between 5 and 60 seconds, and resumes the moment someone speaks again. Active sessions still bill at $0.05 a minute, charged per second, but the rate drops to $0.02 a minute while a session sits in standby, a 60% cut during the quiet stretches.
Why it matters: most voice sessions, phone menus, voice assistants, support lines, spend a good share of their time with nobody talking. Until now that dead air billed at the full rate unless you built your own silence detection to tear the session down and restart it. Standby mode moves that cost-saving into the API itself.