← All posts

GPT-Live-1.1: what's new and is it free?

2026-10-18 · 4 min read

GPT-Live-1.1 is OpenAI's update to the voice model behind ChatGPT Voice. It adds new voices, more natural accents in several languages, smoother handoffs to the model doing the heavy lifting, and fewer of those repeated "checking..." pauses. And yes, it's free in a limited form: Free and Go accounts get a smaller version called GPT-Live-1.1 Quip, while Plus, Pro and Business get the full model.

@OpenAI announced it on October 15 with a four-word pitch, "New voices. Better accents," and said it was rolling out that day to Voice in ChatGPT, Codex and dots, with the API "coming soon." The original GPT-Live launched in July, and we covered that release here. Version 1.1 refines that model rather than replacing it.

What's new in GPT-Live-1.1

About an hour after the launch post, Justin Uberti, who leads realtime AI at OpenAI and created WebRTC, summed it up on X. @juberti listed "much better multilingual support," "more natural delegation," and "improved steerability" as the three headline changes.

Based on OpenAI's announcement and its developer forum post, here's what changed in practice:

  • 13 new voices, two each in French, German, Hindi, Japanese, Korean and Mexican Spanish, plus a new English voice called Bubbie. Which voices you see depends on your region and account.
  • Better accents. OpenAI says pronunciation now sounds more natural across languages. The July launch warned that some languages could come out with a non-native accent, and this update goes after that.
  • Fewer "checking..." loops. When GPT-Live-1 searched the web, it tended to fill the gap with the same word over and over. Version 1.1 varies its short filler phrases, so the conversation sounds less robotic while it waits.
  • Smoother delegation. GPT-Live hands harder questions to a stronger reasoning model in the background. OpenAI says that handoff now feels more natural, though it published no numbers.
  • Better steerability, meaning the model is supposed to follow instructions about tone and style more closely.

On October 16, @mark_k pointed out the second model in the release, GPT-Live-1.1 Quip, and argued that voice will become a more important way to use AI. Not everyone is happy. On OpenAI's developer forum, several users say voices they already knew now sound different.

Is GPT-Live-1.1 free?

Partly. OpenAI's voice usage table, as summarized by DataNorth on October 16, breaks down like this (rolling 24-hour windows):

Plan Model Voice allowance
Free GPT-Live-1.1 Quip Limited, subject to change
Go GPT-Live-1.1 Quip 3 hours
Plus GPT-Live-1.1 3 hours
Business GPT-Live-1.1 3 hours (Standard) or 15 hours (Premium), then credits
Pro $100 GPT-Live-1.1 15 hours
Pro $200 GPT-Live-1.1 Unlimited, with abuse limits

OpenAI hasn't said how much quality Quip gives up next to the full model. Limits like these change often, so check the usage notice in your own account before you plan around them.

You don't have to do anything to turn it on. It arrives on web, iOS and Android automatically. To pick one of the new voices, open Settings, go to Voice and choose from the Voice list. You set the language in the same menu.

What about the API?

Not yet. OpenAI's own wording is "coming soon," and the developer forum announcement gave no date either. GPT-Live-1 has been in the API since September 10, so anyone building a phone agent today is still on the older model. One more thing to know: The Rundown's guide points out that OpenAI's service terms limit ChatGPT Voice output to non-commercial use, so don't record it for an ad.

My read

This is a polish release, and polish is what voice needs right now. When people hang up on a voice agent, it's usually because it paused too long, repeated itself or mangled a name. Raw intelligence is rarely the complaint. Fixing the "checking..." loop and the accents goes straight at the reasons people hang up. Better multilingual voices also matter more than they look for a US business. Plenty of Fox Valley callers would rather talk in Spanish, and Mexican Spanish is one of the six languages that got new voices.

I'd hold off on rebuilding anything until 1.1 reaches the API with a real price. When it does, the gap between a voice agent and a decent human receptionist on routine calls gets a little smaller again.

What this means for a small business

The voices are getting good enough that the harder work sits elsewhere: deciding which calls a machine should take, what it is allowed to promise, and when it hands the caller to a person. If you want to map that out before you buy anything, our free process audit starts with exactly those questions.

08 / Start here

Which of this can you use today?

Tell us what you use today and we'll reply within one business day with what it would take. Free, 20 minutes, no pitch deck.

Email

pgorski@newfacedesign.com

Phone

+1 (773) 627-2176

Based in

Chicago area

Working with clients everywhere

Skip the form: grab a 20 minute slot.

Open the calendar →