OpenAI launched GPT Live. The old ChatGPT voice mode was dumb, this one reportedly isn't

The old ChatGPT voice mode suffered from one main issue – as soon as you switched to voice, you were talking to a significantly dumber version of the model than you were used to from text. GPT Live promises to end this: for more complex queries, it reportedly uses the full-fledged GPT-5.5, and it can also listen and respond simultaneously, not strictly alternately – I speak, you are silent, and vice versa. We tested it in live operation and also found a few things that OpenAI doesn’t mention in its press release.

Why the old voice mode was annoying and what GPT Live changes

The previous ChatGPT Advanced Voice Mode worked on a turn-based principle – the model had to wait for you to finish speaking, and it easily mistook a pause or background noise for the end of a sentence. Hence the annoying interruptions where you literally talked over the assistant. GPT Live is built on a full-duplex architecture – it doesn’t listen to you message by message, but continuously processes your voice and decides several times per second whether to speak, continue listening, wait, or interject only when appropriate.

According to openai.com, there are two versions – the more powerful GPT-Live-1 for paying users and the lighter GPT-Live-1-mini as the default free mode. For more complex queries (deeper reflections, web search), GPT Live in the background uses the GPT-5.5 model and brings the result back into the conversation. It also features “Hey Chat” wake phrases, where the model remains silent and waits until you address it again, and live simultaneous translation – exactly what the official promotional visual shows: three elderly women at a table chatting with a phone as if it were a friend, with GPT Live translating for them instantly. This target group makes perfect sense – older generations are not always best accustomed to new technologies (and may not even want to be), and an assistant who speaks normally and doesn’t interrupt is much more accessible to them than classic chat.

How we got GPT Live running (and why it didn’t activate at first)

Activation wasn’t entirely smooth. We opened the ChatGPT app and tapped the blue Live button in the bottom right – but instead of the new mode, only a small bubble opened at the bottom of the screen. When directly asked if it was already running on the new model, the app honestly admitted that it wasn’t, and that we couldn’t do anything about it ourselves – it’s server-side.

We didn’t give up: in Google Play, we manually searched for and installed all pending updates (the ChatGPT app wasn’t even among them), and most importantly, we restarted the phone. Only after the restart did the app open a large blue-and-white orb across almost the entire screen instead of the small bubble – and when asked if the new voice mode was running, it finally answered yes. By the way, restarting your phone is a good idea even beyond this – it clears any potential malicious software from memory, so it’s a useful general prevention measure.

How many minutes of GPT Live do you get per day?

We asked GPT Live this very question directly, in the middle of a live conversation about how long we would be able to chat with it daily. The answer wasn’t as clear as we had expected.

🎙️ Unofficially: around 60 minutes per day OpenAI has not published any official minute limits for paid plans – only that limits may change continuously and after they are exhausted, the app will switch to a weaker model. According to the experience of early users (not from official documentation), it’s roughly an hour a day in the full version of GPT Live. Take this as an unverified estimate – the model itself admitted it wasn’t 100% sure.

This seemed like quite important information to us – if you want to build a work habit around GPT Live (e.g., dictating notes on long walks), it’s good to know that the app can subtly switch you to a dumber model at any time without explicitly announcing it.

Want to dictate elsewhere? Android can do it too

By the way, we dictated parts of this article – not via GPT Live, but via the desktop app Wispr Flow, which costs a few hundred crowns per month and handles unlimited dictation. If you want to dictate notes for free anywhere, Google Keyboard on Android also does a pretty good job – just tap the recording icon on the keyboard surface. It’s not as smart as GPT Live (it only transcribes what you say, including minor corrections and removal of “hmm” and “uhm”), but it has one annoying flaw: if you turn the phone horizontally or just awkwardly tilt it, the ongoing dictation is canceled without warning. It’s easy to end up talking into thin air for another five minutes.

We live in a time when similar voice tools are appearing every week. Not everything that looks functional is truly functional – and that’s exactly why Svět Androida is here to test it for you.

Have you tried GPT Live? Is it as smart in voice as it is in text?