Edgar Cervantes / Android Authority
TL;DR
- OpenAI’s new voice model — GPT-Live — can listen and speak at the same time.
- For questions that require more thinking or a web search, GPT-Live can keep talking while it processes things in the background.
- GPT-Live-1 and GPT-Live-mini are rolling out now and will be available on the ChatGPT apps and on the web.
Interacting with AI models by talking to them has been possible for a while, but it has never really been the most seamless experience. However, OpenAI’s latest models could change that for the better.
OpenAI has today announced its new GPT-Live voice model. It’s built on a full-duplex architecture, which means the model can listen to users and speak simultaneously. This could result in more natural conversations with GPT-Live. The model will reportedly be able to show it’s paying attention using phrases like “mhmm” or “yeah” while the user is speaking.
To do all this, GPT-Live continuously processes inputs and decides whether to speak, listen, or pause while listening to the user. This way, it can interject with natural responses even while the user continues speaking. An example conversation shared by OpenAI showed GPT-Live engaging with a user in a considerably more natural way than before.
GPT-Live also takes a new approach to questions that require it to think more or search the web. The model basically hands off the task to a model such as GPT-5.5 and continues interacting with the user. It’s a neat way to ensure the conversation doesn’t get stuck while ChatGPT does the research in the background.
GPT-Live is also better at listening to users overall. If a user pauses to think while talking, GPT-Live will wait for them to finish speaking. Users will also be able to ask ChatGPT to stay quiet and listen, and it’s also better at focusing on the user’s voice without getting distracted by ambient noise.
The new model also supports visual responses. It can show rich visual cards for questions about the weather, stocks, sports, and more.
If you’ve used Gemini Live in the past, OpenAI’s new model could appear fairly similar. However, GPT-Live seems to be more advanced than Google’s offering. Gemini Live can keep up with long conversations, but it doesn’t respond quite the same way as GPT-Live does in its demos. GPT-Live can also display visual results, and delegate tasks to other OpenAI models to keep the conversation smooth and natural, which isn’t something Gemini Live can do yet. That could change in the future, but for now, GPT-Live does seem to have the upper hand.
OpenAI is releasing two GPT-Live models: GPT-Live-1 and GPT-Live-mini. Both models are rolling out today. GPT-Live-1 will be the default experience for ChatGPT Pro, Plus, and Go users, while GPT-Live-mini will be the default model for free users. The new voice experience will be available across ChatGPT’s smartphone apps as well as on the web.
Thank you for being part of our community. Read our Comment Policy before posting.
