OpenAI has introduced GPT Live, a new generation of voice models that now power ChatGPT Voice. The company describes GPT Live as its smartest and most natural sounding voice model yet, built to make talking to an AI feel closer to talking to a real person.
What is ChatGPT Voice?
ChatGPT Voice is the feature that lets you speak to ChatGPT out loud instead of typing, and hear it reply back in a spoken voice. Think of it like a phone call with an AI assistant. Until now, this voice feature worked in a fairly robotic, turn based way. You spoke, waited, and then the AI replied. GPT Live changes how this works under the hood, and the result is a conversation that flows far more naturally.
Why did the old voice AI felt clunky?
To understand why GPT Live matters, it helps to know how voice AI worked before. The original ChatGPT Voice used three separate systems working one after another. One system converted your speech to text, another figured out a response, and a third converted that response back into speech. This chain of steps caused delays and often made replies sound slow and stilted.
A later version, called Advanced Voice Mode, combined these steps into a single system, which made replies faster and smoother. However, it still worked in strict turns. The AI had to wait for a person to stop talking before it could respond, and it often struggled to tell the difference between someone pausing to think and someone finishing what they wanted to say. This meant the AI would sometimes interrupt at odd moments.
What does GPT Live do differently?
GPT Live is built on what OpenAI calls a full duplex architecture. In simple terms, this means the model can listen and speak at the same time, much like two people having a real conversation. It can process what a person is saying while also deciding, many times a second, whether to respond, stay quiet, or wait for the person to finish. This allows for small natural cues like saying 'mhmm' while listening, better timing during back and forth exchanges, and the ability to interrupt or be interrupted in a way that feels less awkward.
GPT Live is also designed to handle harder questions without slowing down the conversation. When a query needs a web search, deeper reasoning, or more complex work, GPT Live quietly hands the task off to OpenAI's latest frontier model in the background. At launch, this background model is GPT 5.5. While that heavier model works on the answer, GPT Live keeps the conversation going, so users are not left sitting in silence waiting for a response.
New features users will notice
With GPT Live powering ChatGPT Voice, users can expect more natural conversations, including better handling of interruptions, pauses, and moments where they simply want ChatGPT to listen without jumping in. Voice responses will also be smarter, since users can now choose between Instant, Medium, and High reasoning options depending on how much thought a question needs.
Listening has also improved, with the assistant handling background noise better and waiting through pauses instead of cutting in. Another new addition is visual responses, where ChatGPT Voice can display rich visual cards for things like weather, sports scores, or stock prices while a person is talking, rather than only replying with sound. GPT Live also enables more natural live translation, since the model can process speech continuously rather than waiting for full sentences.
Who can use it?
GPT Live is rolling out globally starting today across iOS, Android, and the ChatGPT website. It comes in two versions. GPT Live 1 will become the default voice model for ChatGPT Go, Plus, and Pro subscribers, while a smaller version called GPT Live 1 mini will become the default for users on the free plan. OpenAI has said it plans to bring GPT Live to its API soon, which would allow developers to build the same voice technology into their own apps.
