GPT-Live-1.1 is OpenAI’s updated voice model for ChatGPT. It adds 13 new voices, better accents in more languages, and fewer awkward “checking…” pauses. It started rolling out on October 15 to all users, but how long you can talk depends on your plan.
Here is the short version before the details.
- Rollout began October 15, 2026, per OpenAI’s developer community announcement.
- 13 new voices: one English voice (Bubbie) and two each in French, German, Hindi, Japanese, Korean and Mexican Spanish.
- It powers voice in ChatGPT, Codex and the dots app.
- An API release is “coming soon,” with no date yet.
- Free users get a smaller model called Quip with limited access.
If you mostly talk to ChatGPT in English, the change may feel small. If you use it in another language, it’s the bigger story.
What GPT-Live-1.1 changes in ChatGPT Voice
OpenAI says GPT-Live-1.1 improves non-English speech, makes accents sound more natural, and says “checking…” less often while it looks something up. It also says delegation feels more natural and that you can steer a conversation more easily mid-sentence.
That “checking…” fix matters more than it sounds. In the older model, a web search could trigger the same filler line again and again. Now the wording should vary and appear less often, according to Business Upturn.
The update runs on two models. GPT-Live-1.1 is the full version, and GPT-Live-1.1 Quip is the smaller one. Quip replaces the old mini model for people on lower tiers.
OpenAI hasn’t published benchmarks or latency numbers for the new model. So “better” is OpenAI’s claim for now, not something we can measure.
The 13 new voices, language by language
English speakers get just one new voice and keep the nine existing ones. Almost all the new options are for other languages.
| Language | New voices | Who benefits |
|---|---|---|
| English | 1 (Bubbie) | Anyone wanting a fresh option |
| French | 2 | Native speakers and learners |
| German | 2 | Native speakers and learners |
| Hindi | 2 | Hindi speakers, a language OpenAI called out specifically |
| Japanese | 2 | Native speakers and learners |
| Korean | 2 | Native speakers and learners |
| Mexican Spanish | 2 | Users who want a regional accent |
Only Bubbie is named in the sources we found, so we can’t tell you what the other 12 are called. OpenAI also lets you create a voice from a text description or an approved recording, according to an October 16 update in its developer community thread.
How much voice time you get on each plan
Everyone can use the update, but time limits differ. OpenAI’s announcement doesn’t list them, so these figures come from Business Upturn and may change.
| Plan | Model | Voice time |
|---|---|---|
| Free | GPT-Live-1.1 Quip | Limited access |
| Go | GPT-Live-1.1 Quip | 3 hours |
| Plus | GPT-Live-1.1 | 3 hours |
| Pro ($100) | GPT-Live-1.1 | Up to 15 hours |
| Pro ($200) | GPT-Live-1.1 | Unlimited |
Time limits are the catch to watch. If you talk to ChatGPT during a long commute, three hours can disappear fast. Pro users who run several tasks at once may also like the ChatGPT parallel agents upgrade that landed the same week.
Some users say their old voices sound different
Not everyone is happy. In the same developer thread, several users say familiar voices such as Sol, Spruce, Arbor and Maple now sound deeper, more generic, or oddly accented. One reported a slip into British English.
These are unofficial reports, and OpenAI hasn’t responded to them in the thread. A few people also say standard voice mode isn’t selectable on their laptop, only Live. Treat that as anecdotal until OpenAI confirms it.
There is also a safety note. According to The Neuron, OpenAI’s synthetic tests showed statistically significant regressions in areas such as sexual content, mental health and emotional reliance, while real-world audio tests showed no significant regressions. Neither set tells you how often harm happens in daily use.
Also, spoken approval doesn’t work for actions that need a go-ahead. You still have to confirm those on screen.
When the GPT-Live-1.1 API arrives
OpenAI says API access is coming soon but hasn’t given a date. A staff reply confirmed it wasn’t live yet.
The last launch offers a clue. GPT-Live-1 reached ChatGPT on July 8 and the API on September 10, about two months later. That’s a pattern, not a promise. If you build voice apps, plan for a wait, and see how Grok Voice 3.0 prices its API in the meantime.
Frequently Asked Questions
Is GPT-Live-1.1 free?
Partly. Free users get limited access through the smaller Quip model. The full model comes with Plus and above, according to Business Upturn.
How do I get the new ChatGPT voices?
Update the ChatGPT app and open Voice. OpenAI says the rollout reached all users on October 15, but availability can vary by account and region, so they may not all show up yet.
Does GPT-Live-1.1 work in other languages?
Yes. OpenAI says it improved French, German, Hindi, Mexican Spanish, Korean, Japanese and many others. It hasn’t published accuracy numbers for any language.
Is there a GPT-Live-1.1 API?
Not yet. OpenAI lists it as coming soon with no release date.
Why does my favorite ChatGPT voice sound different?
Several users report the same thing after the update. OpenAI hasn’t explained it, so try switching voices or updating the app.
My take: try the update if you speak French, German, Hindi, Japanese, Korean or Spanish, because that’s where the gains are claimed. English users can skip the hunt for Bubbie. And if you rely on a voice you love, check it still sounds right before your next long call.
Leave a Reply