GPT-Live-1.1 in ChatGPT: new voices, more natural accents and what changes for creators

What GPT-Live-1.1 is and what changes

GPT-Live-1.1 is the update to the model behind Live, the ChatGPT Voice mode that listens and speaks at the same time, so you can interrupt it mid-answer as you would on a phone call. According to the official ChatGPT release notes, starting October 15, 2026 OpenAI is rolling out GPT-Live-1.1 and GPT-Live-1.1 Quip, a mini version of the model.

OpenAI lists three changes:

• more natural accents across languages;
• new voices for French, German, Hindi, Japanese, Korean and Mexican Spanish, plus an English voice called Bubbie;
• a wider range of short phrases to let you know it is searching the web, instead of a repetitive "checking…".

In the announcement on the OpenAI Developer Community the update is described as covering ChatGPT, Codex and dots, with "13 new voices" and better performance in French, German, Hindi, Mexican Spanish, Korean, Japanese and "many other non-English languages". OpenAI adds that delegation is more natural and steerability has improved. It has not published numbers or benchmarks for these improvements.

The new ChatGPT Voice voices, language by language

You pick a voice in Settings → Voice → Voice. OpenAI notes that the voices you see may vary by region and account, and the rollout is gradual: if they are not there yet, your account has not been updated. These are the additional Live voices listed on the ChatGPT Voice help page as of October 20, 2026:

The first seven rows are the 13 voices announced on October 15; the help page now also lists two voices for Brazil. The nine core voices remain: Arbor, Breeze, Cove, Ember, Juniper, Maple, Sol, Spruce and Vale.

For any other language, set the one you speak most often in Settings → Voice → Language so ChatGPT understands you better, or ask it to switch language during the conversation. Some users on the forum report that familiar voices such as Sol and Spruce sound different after the update: if you rely on one for your content, listen again before recording.

How much you can use it: limits by plan

The model you get depends on your plan: Free and Go get Quip, higher tiers get the full model. Limits are measured over a rolling 24-hour period, according to the official ChatGPT Voice page:

What Live can do: search the web, use memory, show visual results, work with text and images in the same chat and use the plugins connected to your account. What it cannot do: video and screen sharing, which stay in Advanced mode on the iOS and Android apps. While it speaks, responses also appear as text in the chat.

On data: audio clips from Live conversations are stored with the chat for 30 days. OpenAI says it does not use them to train its models unless you turn on audio sharing in Settings → Data Controls.

In the API: no GPT-Live-1.1 yet, but voices from a description

If you build apps, assistants or talking avatars, the short answer is: GPT-Live-1.1 is not in the API yet. The announcement says "Coming soon to the API", with no date. The API still runs GPT-Live-1, available since September 10, 2026 and, according to Tech Times, billed at $0.05 per minute for the voice layer.

In the same days, however, the OpenAI API changelog added two concrete features for GPT-Live-1:

• October 16, 2026, prompt-to-voice: every API customer can create a custom voice from a text description (accent, age, tone, pace) and reuse it in GPT-Live sessions. Creating the voice is free; speech is billed at regular session rates. The same day brought nine new API voices: Nuri, Juni and Flitz (German), Brise and Sillage (French), Nira (Hindi), Harema and Shida (Japanese), Noeul (Korean).
• October 15, 2026, native Decisions integration: GPT-Live can call the Decisions API directly while the user is still speaking, without a round trip through your app.

The Custom voices guide sets clear rules: up to 100 prompt-generated voices per organization, adult voices only, no imitation of real people or characters, and you must tell listeners the speech is AI-generated. Voices created from a recording require a consent recording and a reference sample.

What it means for you as a creator using AI voice and avatars

GPT-Live-1.1 is a conversation tool, not a voiceover generator: it does not export audio files ready for editing. For a creator it is useful at other points in the workflow:

1. Rehearse scripts out loud. Read the script for a reel and ask Live to interrupt you where the pacing drops: listening and answering at the same time is exactly what it is built for.
2. Prepare content in more languages. If you also publish in French, German or Spanish, ask Live how a line sounds with a more natural accent than before, then pick the right voice for your avatar.
3. If you build a voice assistant, prompt-to-voice gives you a branded voice without recording anyone. Keep in mind that OpenAI says voices are not guaranteed to be unique: others may generate similar ones.

For videos with your own face and real voice, cloning is still the way to go: see the options in our AI clone guide and the HeyGen vs Synthesia comparison. To do it in half an hour with a guided method, there is Create your AI clone in 30 minutes. And if you follow the race toward personal assistants, read about Gemini 4 Argon and Amazon Bee.