GPT-Live: ChatGPT can now listen and talk at the same time

What's new

OpenAI has launched GPT-Live, a new generation of voice models for ChatGPT Voice. The rollout began on July 8, 2026. There are two versions: GPT-Live-1 and the smaller GPT-Live-1 mini.

Free users get GPT-Live-1 mini by default. Subscribers on Go, Plus and Pro get GPT-Live-1. According to the sources, free users can probably only use the mini model to a limited extent within a rolling 24-hour period, but the specific limit is not stated.

GPT-Live is rolling out globally on iOS, Android and ChatGPT.com. According to ETV Bharat, it also works with Apple CarPlay. OpenAI expects all users to have it within a few days. The sources mention no special restrictions for the EU or Denmark.

What's smart

The biggest change is that the model listens and speaks at the same time (the technical term is full-duplex). The old voice mode was turn-based: it waited until you paused before it answered. Now the conversation feels more like an ordinary phone call. The model can interrupt itself, make small acknowledging sounds and handle longer pauses without answering too early. It can also translate live.

For demanding tasks such as web search and deep thinking, the model hands the work off to GPT-5.5 in the background while the conversation continues. You can choose between three levels of thinking yourself: Instant, Medium and High. During the conversation, visual cards can also be shown, for example about weather, stocks, sports and maps.

OpenAI has also updated the nine built-in voices, which cannot imitate real people. The model is tuned to listen better when there is background noise. You open the feature by tapping the voice button in the ChatGPT app, and it appears when it reaches your account.

Sam Altman, CEO of OpenAI, says the model feels magical and "real." He has always preferred typing to talking with an AI, but now believes that is changing.

Cheaper or better?

OpenAI says GPT-Live performed better than the previous Advanced Voice Mode on conversational flow and on tests of reasoning and search. These are OpenAI's own tests, and the sources do not include specific figures.

Subscription prices are not stated in the material, and there is no comparison with competitors' voice features. The good news for your wallet is that one version is included in the free edition, but free users can probably only use it to a limited extent within a rolling 24-hour period. The specific limit is not stated. OpenAI plans to bring GPT-Live to its API (the access point for developers), but no timeline has been given.

There are also limitations. GPT-Live is optimized for some of the most widely used languages in ChatGPT. In other languages, there may be an accent or less fluent speech. It is not stated whether Danish is among the optimized languages. At launch, neither video alongside voice nor screen sharing is supported. The older voice modes can still be used for that, and OpenAI is working on adding the features.

What it's good for

GPT-Live suits anyone who would rather talk than type: questions on the go when your hands are busy, or a chat about an idea where you want to be able to interrupt and correct along the way. The ability to handle background noise and pauses makes it more usable outside quiet rooms.

Live translation can help when traveling or in meetings with people who speak another language. Keep an eye on how well Danish works, though. Users who need video or screen sharing alongside voice will have to stick with the older voice modes for now.

OpenAI has introduced special safety checks for voice conversations about self-harm, violence and sexual content. There is protection for teenagers and parental controls, and parents are notified in cases of higher risk of self-harm.