ChatGPT Can Now Talk and Listen at the Same Time, OpenAI Says

OpenAI has rolled out a new voice system for ChatGPT that lets the AI listen and speak at the same time, doing away with the stop-and-start rhythm that has defined AI conversations for years. The company calls it GPT-Live, and it's now replacing Advanced Voice Mode for users everywhere.

The change sounds technical, but the effect is simple: no more waiting. Older voice assistants had to detect a pause, decide the person was finished talking, then generate a reply — a process that often mistook a breath or background noise for the end of a sentence and cut people off mid-thought. GPT-Live processes audio continuously instead, making decisions about whether to speak, stay silent, or interrupt many times per second. It can slip in a quick "mhmm" while someone's still talking, the way a person would.

Two versions are shipping. GPT-Live-1 is the default for paying ChatGPT subscribers, while a lighter GPT-Live-1 mini serves free users. Both are live now on iOS, Android and the web.

For anything that requires real thinking — a web search, a multi-step question, actual reasoning — GPT-Live quietly hands the task off to GPT-5.5 running in the background, then folds the answer back into the conversation without breaking its flow. OpenAI is positioning this as a first step toward voice becoming a genuine interface for longer, more autonomous work, not just a novelty for asking about the weather.

That shift toward listening in real time raises the obvious question of what happens to what's said. OpenAI states that GPT-Live relies only on a fixed set of pre-approved voices and includes safeguards meant to stop the system from mimicking a real person's voice. The company also says it built dedicated safety testing for the voice format itself, covering scenarios involving emotional dependence on the assistant, self-harm, and mental health crises, with real-time responses designed to redirect a conversation and surface support resources when needed. Parents can restrict teen access to voice chat entirely through parental controls.

In OpenAI's own head-to-head testing against the previous voice mode, people preferred GPT-Live-1 roughly three-quarters of the time. The mini version, aimed at free users, still won out over the old system in about seven of ten comparisons.

Not everything is finished. Live translation demos have shown an accent creeping into other languages, and OpenAI hasn't said which languages are fully supported. Video and screen-sharing during voice chats aren't available yet, either. The company says an API version for developers is coming, though no date has been set — for now, interested businesses can only join a waitlist.

More than 150 million people already use ChatGPT's voice features weekly, according to OpenAI, a number that's likely to climb once GPT-Live becomes the assistant most people are actually talking to.