Back home

GPT-Live-1.1

x+web 0.95
Visit source

Intro

GPT-Live-1.1 is OpenAI’s full-duplex voice model for realtime conversations. The model ID is gpt-live-1.1. At 2026-11-20T04:12:00Z (12:12 Beijing time), OpenAI’s developer community posted GPT-Live-1.1 has arrived in the API, saying it is now available in the API for voice agents with more natural, consistent accents across languages, better handling of precise details such as letters and numbers, and more reliable instruction following. The official model page, GPT-Live-1.1, describes a model that can listen while it speaks.

The same announcement states that GPT-Live-1.1 has powered voice in ChatGPT Plus and Pro since 15 October 2026, and that developers can now use the same model ID in their own apps. The OpenAI API changelog lists “Model: gpt-live-1.1 / API: v1/live/sessions” under the Nov 19 entry in November 2026, and says you can set gpt-live-1.1 in an existing Live session configuration.

This entry is for the developer-facing API launch of the name GPT-Live-1.1, together with the standby billing option that official docs apply to both GPT-Live-1 and GPT-Live-1.1. It is not the already-filed Live 3 or Build Live Apps, and it is not the predecessor GPT-Live-1. Earlier ChatGPT voice support is not the same event as this API release.

Advertisement

Explain

According to the official model page, GPT-Live-1.1 is a full-duplex voice model for multilingual conversations: input and output both include audio and text, and the default snapshot is gpt-live-1.1. Compared with GPT-Live-1, OpenAI says it transcribes speech more accurately and keeps a more consistent accent and voice across languages. You can describe in plain language when the model should hand a task to your backend for reasoning or tool use, without breaking the conversation. It uses the Live endpoint v1/live/sessions. The official model page lists Chat Completions, Responses, Realtime, and several other endpoints as unsupported.

Standby, announced at the same time, is a new GPT-Live API setting that official docs apply to WebRTC sessions for both GPT-Live-1 and GPT-Live-1.1. After an idle timeout between 5 and 60 seconds, inference pauses while the call and conversation history stay open, then resumes when someone speaks. Official pricing is $0.05 per minute while active and $0.02 per minute in standby; backend model and tool usage is billed separately. GPT-Live-1.1 and GPT-Live-1 have separate concurrent-session limits. Mute, standby, and hanging up are different states.

The changelog lists the model under 19 November; the community announcement was posted at 04:12 UTC on 20 November. This entry follows those official pages and does not treat third-party benchmarks or unstated regional availability as facts.

Usage

Start from the official announcement GPT-Live-1.1 has arrived in the API, the model page GPT-Live-1.1, and the API changelog. Set the model to gpt-live-1.1 in an existing Live session configuration and create sessions with v1/live/sessions. For standby, follow the official docs to set a 5–60 second idle timeout on WebRTC sessions. Official pricing is $0.05 per minute active and $0.02 per minute in standby; backend models and tools are billed separately. Muting the microphone or closing the session is not standby.

This is for full-duplex voice agents that can be interrupted and can hand hard work to a backend model mid-call. Voice inside ChatGPT Plus and Pro has used this model since 15 October; that product path is not the same as this API launch. Prices, limits, and regional availability are whatever the official docs say at the time you integrate. This page only restates those official sources. It is not the already-filed Live 3 or Build Live Apps.