OpenAI Launches GPT-Live, a New Full-Duplex Voice Mode for ChatGPT

OpenAI is introducing a new generation of ChatGPT's voice mode called GPT-Live. Instead of taking turns speaking like previous assistants, the model listens and talks at the same time, which is meant to make conversations with the app feel more like talking to a person.

How full duplex works

Previous voice modes in ChatGPT and rival assistants worked in turns: the user would speak, the model would process the input, and then respond. GPT-Live breaks that pattern, since the model can simultaneously receive audio and generate its own spoken response. It can be interrupted mid-sentence and will smoothly adapt to a change of topic instead of finishing a previously planned reply.

OpenAI notes that the model can also stay silent when the other person needs a moment to think, and can interject brief acknowledgments, like a murmur, that in human conversation signal attention without taking over the floor. These are details that used to belong to the world of telephony voice-interface design and are now making their way into a popular AI app.

GPT-Live can listen and talk at the same time. The result is a voice experience that feels remarkably easy to talk to. - from OpenAI's official announcement

Two model variants

Two versions are available at launch: GPT-Live-1 for subscribers on the paid Go, Plus, and Pro plans, and GPT-Live-1-mini as the default choice for free accounts. Behind the scenes, for questions that require web search, deeper reasoning, or more complex work, GPT-Live hands the task off to OpenAI's latest model running in the background, and the result returns to the conversation without breaking its natural rhythm.

The new voice mode arrives alongside a broader rollout of the GPT-5.6 model, which OpenAI plans to release on July 9 following an additional safety review. Together, the two rollouts show that OpenAI is now betting on voice conversation as the primary channel for interacting with ChatGPT, competing directly with voice assistants on smartphones and in cars.

CarPlay without full control

Integration with Apple CarPlay is one of the more practical changes for drivers. ChatGPT with the new voice mode works within Apple's in-car interface, but with notable limitations: the app cannot control vehicle functions, and launching the assistant with a wake word, as with Siri, is not possible due to Apple's rules on third-party voice app integrations. Users must therefore start a conversation manually from the screen.

For Polish ChatGPT users, the new voice mode mainly means a faster, more natural way to use the assistant on their phone and in the car, though at launch without a confirmed Polish voice localization in OpenAI's materials. It's also worth noting that video and screen-sharing features, available in the old voice mode, have not yet made it to GPT-Live and require switching back to the previous system.

Competition in the AI voice assistant segment is growing alongside OpenAI's ambitions. Google is developing its own voice features in Gemini, while Amazon and Apple are working on refreshes of their assistants, Alexa and Siri. Full duplex, if it holds up in everyday use, could become the new standard that other players will have to catch up to.

Sources: 9to5Mac (9to5mac.com), Engadget (engadget.com), CNBC (cnbc.com)