OpenAI GPT-Live-1.1 Cuts Idle Voice Agent Costs by 60%
By Ganesh Karki
Key facts
- Section
- Tech & AI
- Country
- Global
- Published
- 22 Nov 2026
OpenAI GPT-Live-1.1 arrived in the API on 19 November, with a standby mode that cuts the cost of an idle voice session by 60%. GPT-Live-1.1 is a full-duplex voice model that listens while it speaks, and idle WebRTC time now bills at $0.02 per minute instead of $0.05 per minute for active inference.
| Organisation | OpenAI |
How OpenAI GPT-Live-1.1 standby works
Standby is a billing state for WebRTC sessions. Developers set a wake policy of speech, so inference waits for voice activity, and an idle timeout of 5 to 60 seconds, after which the session pauses inference while the WebRTC connection, the session ID and the conversation history stay open. When the user speaks, the session resumes. OpenAI sends session.paused when standby starts and session.resumed when inference restarts. The voice activity threshold runs from 0 to 1 and defaults to 0.5. Backend models and tools that GPT-Live calls through delegation are billed separately and do not change with standby. GPT-Live-1.1 also improves pronunciation, accent consistency and transcription in multilingual voice conversations, and can delegate reasoning and tool calls to a backend model while it keeps talking. Readers can also read our report on OpenAI GPT-6.2 Astra and Sol launch, and more AI news in our Tech section.
What it means
A voice session spends much of its life saying nothing, and until now that silence was billed at the active rate simply because the connection was open. Standby turns silence into something a developer manages, with a price, a timeout and events to count. It fits support calls, form filling, coaching and in-car assistants, where pauses are part of the conversation. The same model has powered ChatGPT Plus and Pro voice since 15 October, so the API release brings a model already in use to developers building their own voice agents.
What the sources do not say
OpenAI has not published benchmark numbers for the transcription or reliability gains in GPT-Live-1.1, so those improvements are vendor claims until developers measure them. The sources do not say how much slower the first reply after waking from standby may be, beyond noting that OpenAI warns it may be slower after a short utterance. They also do not say whether standby will come to WebSocket sessions, which are not covered by the new billing state.
Sources: OpenAI API changelog, the 19 November GPT-Live-1.1 entry and the standby configuration; MarkTechPost, the $0.02 standby rate, the 5 to 60 second timeout and the slower first reply warning; AI Weekly, the API availability and the standby pricing.
Frequently asked questions
What does GPT-Live-1.1 standby cost?
Does standby work on WebSocket sessions?
What changes when upgrading from GPT-Live-1?
Sources and references
Information checked: 22 November 2026
Related stories
Science Corporation Raises $400M for Vision Chip PRIMA
Science Corporation raised $400 million in Series D funding to expand its PRIMA vision-restoration system in Europe and prepare a US launch.
22 November 2026US Smartphone Shipments Fall 9% as Apple Hits 57% Share
US smartphone shipments fell 9% in Q3 2026, as memory constraints hit low-end Android. Apple was the only maker to grow, reaching a 57% share.
22 November 2026Claude Solves Two Black Hole Physics Problems, Anthropic Says
Claude black hole physics results solved two open problems, Anthropic said on 17 November, finishing a 2003 calculation and finding an exact limit.
22 November 2026Apple Creator Studio Adds AI Assistants Across Final Cut and Logic
The Apple Creator Studio update adds AI assistants across Final Cut Pro, Logic Pro, Pixelmator Pro and iWork, rolling out from 18 November.