On October 15, 2026 (U.S. time), OpenAI updated the AI model used for ChatGPT's voice conversations, ChatGPT Voice, and other features to GPT-Live-1.1. Starting that day, it is rolling out gradually to all users. The main changes are new voices and more natural speech in languages other than English.
What's new in GPT-Live-1.1
According to an announcement on OpenAI's developer community, the main changes are as follows.
- New voices: 13 voices added
- Improvements for non-English languages: performance has improved significantly in many languages other than English, including French, German, Hindi, Mexican Spanish, Korean and Japanese
- Fewer repeated stock phrases: it less often inserts the same phrases, such as "Let me check...", over and over
- Handing off tasks and following instructions: the flow of handing work off to other processes is more natural, and it follows instructions more readily
The announcement does not explain in detail which languages the 13 new voices are for. The Korean outlet Digital Today reported that two new voices were added for Korean. We could not confirm whether there are new voices for Japanese.
Where it is available
According to OpenAI, GPT-Live-1.1 is rolling out gradually to all users through updates to ChatGPT, Codex and dots. Codex is OpenAI's coding agent.
A version developers can use through the API is described as "coming soon," with no timing given.
Related changes for developers
On the same day, October 15, OpenAI announced in its developer changelog that gpt-live-1, the API version of its voice model, now supports Decisions, a mechanism that connects the model to an app's processing. It allows the app to carry out actions (such as adjusting the temperature in a car) even while the user is still speaking, or to act in step with what the model says.
Reported bugs
In replies to the announcement on the official community, users have reported problems such as playback in a different voice from the one they selected, the new mode being used even when they choose the standard voice mode, and voices they had used for a long time disappearing. OpenAI staff have replied in the thread, but about the timing of API availability, and as of this writing we have not been able to confirm any response to these bugs.
CoAI's take
GPT-Live-1.1 is an update that improves the range of voices and the naturalness of speech rather than adding new features. Since Japanese is among the languages targeted for improvement, people who talk to ChatGPT by voice may notice a difference. The day before, SpaceXSI (formerly xAI) released Grok Voice 3.0, a voice model for developers, and companies are continuing to improve their voice AI.
However, OpenAI has not published any performance evaluation figures. Bugs with voice selection have also been reported, so if you are concerned, it is best to update the app to the latest version and wait and see for a while.
