Skip to content
NLEN
Illustration: OpenAI pipeline found 948 Linux root bugs in one run: what autonomous bug hunting means

GPT-Live-1.1 in the API: a voice model with cheaper standby

By Ivo Donker · — text: Claude (AI)

What is GPT-Live-1.1? It is OpenAI's voice model that can listen and speak at the same time, and it is now available in the API under the ID gpt-live-1.1. It speaks with more natural and consistent accents across languages and understands letters and numbers more precisely. New is a standby mode: after 5 to 60 seconds of silence the model pauses, and that time costs 2 cents per minute instead of 5 cents.

We read the announcement on the OpenAI developer forum, the model page and the API changelog, which dates the launch to November 19. The news spread through a post by @OpenAIDevs on X and was in the trending briefing of November 20.

What happened

OpenAI is opening GPT-Live-1.1 to developers. According to OpenAI, the model has powered voice in ChatGPT Plus and Pro since October 15. Developers can now use it in their own apps through the Live sessions endpoint. If you already run a GPT-Live-1 session, the changelog says you only swap in the new model name.

GPT-Live is a full-duplex model: it can listen while it speaks, so a user can interrupt it as in a normal conversation. It does not have to do complex work itself. In plain language you describe when the model should hand a task to another model or to your own systems, for reasoning or tool calls, while the conversation keeps going.

Prices and standby

Prices and standby
ItemGPT-Live-1.1
Model IDgpt-live-1.1
Active conversation0.05 dollars per minute
Standby (WebRTC, inference paused)0.02 dollars per minute
Standby timeoutconfigurable from 5 to 60 seconds
Backend model and toolsbilled separately
Standby also forGPT-Live-1

Standby works like this: after the set period of silence the model pauses, but the WebRTC connection stays open. As soon as someone speaks again, the conversation resumes. A rough sum: a ten-minute call with six minutes of silence costs 50 cents without standby. With standby it is 4 active minutes at 5 cents plus 6 minutes at 2 cents, 32 cents in total. The seconds before the timeout kicks in still count as active, so the real saving is a little smaller.

What this means

Voice agents with a lot of silence get noticeably cheaper. Think of a support line where someone looks something up, a kiosk waiting for the next question or an assistant listening in on a meeting. Until now you paid the full rate for silence. The bill still has two parts: the GPT-Live minutes and the cost of the model you hand tasks to, for example GPT-6.2 Astra or Sol.

Splitting work between a fast voice model and a heavier backend model is a familiar pattern in agentic AI: the voice model keeps the conversation going while the other model does the thinking. Perplexity also adds voice to its agents, as in Teammates. The promise of more consistent accents matters for non-English apps, but test it yourself: OpenAI gives no measurements per language.

What we did not check

We did not test how well the model speaks or understands Dutch or other languages. The announcement gives no figures for the improved pronunciation or transcription. According to the news site AI Weekly, the API includes 38 voices, seven of them new; we did not find that in OpenAI's documentation. We did not open the full pricing page.

Checked on November 20, 2026.

This article was written with AI (Claude) and checked against the sources below.

Sources