OpenAI Releases GPT-Live-1.1 for the Live API (2026)

Written by

in

Takeaway: On November 19, 2026, OpenAI’s API changelog added GPT-Live-1.1 for the Live API. The company says the model improves pronunciation, accent consistency, and transcription in multilingual voice conversations. Developers use the model ID gpt-live-1.1 with the existing v1/live/sessions endpoint. Voice sessions are billed by duration at $0.05 per active minute and $0.02 per standby minute; backend model and tool usage is billed separately. This is an API model update, not a ChatGPT product feature, and it is distinct from the already covered ChatGPT Live Dictation rollout.

The details below come from that November 19 changelog entry, the official GPT-Live-1.1 model page, the Live session management guide, the voice cost guide, and the API pricing page. Official sources do not say GPT-Live-1.1 is available in the ChatGPT app or on Plus or Pro plans, and this article does not add that claim. The November 19 entry also does not publish a new language or voice list for 1.1.

What GPT-Live-1.1 is

The model page describes GPT-Live-1.1 as a full-duplex voice model for multilingual conversations. The model ID is gpt-live-1.1, which is also the default snapshot. OpenAI says GPT-Live-1.1 can listen while it speaks. Compared with GPT-Live-1, it transcribes speech more accurately and maintains a more consistent accent and voice across languages. The changelog and model page both tell developers to describe, in plain language, when the model should hand a task to a backend for reasoning or tool use.

Input and output modalities are audio and text. Supported features listed on the model page are streaming and function calling. The changelog says you can use gpt-live-1.1 in an existing Live session configuration. Official docs also say to choose the model when you create the session, and to start a new session if you need to change the model later.

This is not ChatGPT Live Dictation

The similar “Live” label is easy to mix up with two other items already on this site. ChatGPT Live Dictation, announced in the November 18 ChatGPT release notes, is a composer feature that turns speech into editable text as you talk. It is a ChatGPT product change, not an API model release.

Separately, OpenAI is still retiring Advanced and Standard Voice from ChatGPT Voice on December 11, 2026, with remaining users moving to Live voice. That retirement is a change to the ChatGPT Voice product. The November 19 changelog presents GPT-Live-1.1 as an update to the Live API model used through v1/live/sessions. Official ChatGPT release notes are not the source for this article, and they are not cited here as making GPT-Live-1.1 available in the ChatGPT apps.

The Live API, not the Realtime API

The model page lists Live v1/live/sessions as the supported endpoint. Chat Completions, Responses, the Realtime API, Realtime translation, Realtime transcription, Assistants, Batch, fine-tuning, embeddings, image and video endpoints, speech generation, transcription, translation, Moderation, and legacy Completions are listed as not supported. GPT-Live-1.1 is not a drop-in model for those APIs, and it should not be confused with the Realtime API.

The same November changelog earlier listed GPT-6.2 Astra and Sol for the Responses and Chat Completions APIs. Those are separate models on different endpoints. GPT-Live-1.1 stays on the Live sessions API.

Rate limits are also model-specific. The model page says GPT-Live-1.1 and GPT-Live-1 have separate concurrent-session limits, and that each model’s limit applies across inference regions. OpenAI points readers to their organization’s limits rather than publishing a single public number, and this article does not invent one.

How sessions are billed

The model page and the API pricing page agree on the duration rates. Voice sessions are billed by duration. Backend model and tool usage is billed separately. For both gpt-live-1.1 and gpt-live-1, the official rates are $0.05 per minute while a session is active and $0.02 per minute while it is on standby. The pricing page adds that GPT-Live bills voice sessions per second.

The cost guide explains what those two states mean. Active time includes speech, silence before the idle timeout expires, and time spent waiting for backend work while inference is running. The standby rate applies while inference is paused in a WebRTC session that has been configured for standby. The November 19 changelog also notes that developers can configure standby for GPT-Live-1 and GPT-Live-1.1 WebRTC sessions to pause inference during inactivity and resume when someone speaks. Muting the microphone leaves the session active. When the conversation is finished, close the session and collect its final usage.

With the default startup behavior, creating a WebRTC session with POST /v1/live/sessions bills 15 seconds of voice duration while the session initializes. The cost guide says that amount is credited against duration charges once the session starts running, so you should not add another 15 seconds when estimating. A 90-second active session already includes the 15 seconds billed at initialization. The same guide’s example is straightforward: 60 seconds active and 30 seconds on standby is one minute at the active rate plus half a minute at the standby rate, plus any backend charges.

Key facts

ItemWhat official sources say
DateNovember 19, 2026, in the OpenAI API changelog.
ModelGPT-Live-1.1, model ID and default snapshot gpt-live-1.1. Full-duplex voice model for multilingual conversations.
What changed versus GPT-Live-1Improved pronunciation, accent consistency, and transcription. The model can listen while it speaks and, compared with GPT-Live-1, transcribes more accurately and keeps a more consistent accent and voice across languages.
EndpointLive v1/live/sessions is supported. Chat Completions, Responses, Realtime, and the other endpoints listed on the model page are not.
FeaturesStreaming and function calling. Input and output modalities: audio and text.
Duration pricing$0.05 per active minute and $0.02 per standby minute, for both GPT-Live-1.1 and GPT-Live-1. Sessions are billed per second. Backend model and tool usage is billed separately.
StandbyAvailable for GPT-Live-1 and GPT-Live-1.1 WebRTC sessions. Muting the microphone leaves the session active. Close the session when finished.
ChatGPT app availability, languages, and voicesOfficial API docs do not state ChatGPT app, Plus, or Pro availability for GPT-Live-1.1. The November 19 changelog does not publish a new language or voice list.

FAQ

What is GPT-Live-1.1?

It is an updated Live API voice model announced in OpenAI’s November 19, 2026 API changelog. The model ID is gpt-live-1.1. Official docs describe it as a full-duplex model for multilingual conversations that improves pronunciation, accent consistency, and transcription compared with GPT-Live-1.

Is GPT-Live-1.1 the same as ChatGPT Live Dictation?

No. ChatGPT Live Dictation, announced on November 18, is a ChatGPT composer feature that turns speech into editable text. GPT-Live-1.1 is an API model used through v1/live/sessions. It is also separate from the December 11 retirement of Advanced and Standard Voice in ChatGPT Voice.

How much does a GPT-Live-1.1 session cost?

Official duration pricing is $0.05 per active minute and $0.02 per standby minute. The pricing page says GPT-Live bills voice sessions per second. Backend model and tool usage is billed separately, so a complete estimate needs those charges as well.

What is standby billing?

The lower $0.02-per-minute rate applies while inference is paused in a WebRTC session configured for standby. The changelog says standby is available for GPT-Live-1 and GPT-Live-1.1. Active time still includes speech, silence before the idle timeout, and waiting for backend work while inference is running. Muting the microphone leaves the session active.

Which API endpoint does GPT-Live-1.1 use?

The official model page lists Live v1/live/sessions as supported. It lists Chat Completions, Responses, the Realtime API, and a long list of other endpoints as not supported. GPT-Live-1.1 should not be confused with the Realtime API.

Is GPT-Live-1.1 available in the ChatGPT app?

Official API changelog and model documentation do not state ChatGPT app, Plus, or Pro availability for GPT-Live-1.1. This article does not infer that the API model is rolling out in those products.

Which languages and voices does GPT-Live-1.1 add?

The November 19 changelog does not publish a new language list or a new set of voices for GPT-Live-1.1. Official text describes improved pronunciation, accent consistency, and multilingual transcription. This article does not invent a 1.1 language or voice list.

Do GPT-Live-1.1 and GPT-Live-1 share rate limits?

No. The model page says the two models have separate concurrent-session limits, and that each model’s limit applies across inference regions. Check your organization’s limits for the numbers that apply to your account; official docs do not publish a single public cap here.

Comments

Leave a Reply