OpenAI Gives ChatGPT a Voice That Listens and Talks at the Same Time

OpenAI is rolling out a major upgrade to ChatGPT’s voice capabilities, introducing two new models — GPT-Live-1 and a smaller version — that can now listen and speak simultaneously. The update, available to all users on mobile and web, aims to make conversations with the AI feel more natural and less robotic.

The key technical shift is something called continuous interaction, which lets the AI process incoming audio while it’s still generating speech. That means you can interrupt it mid-sentence, or ask a new question while it’s still finishing a thought. It also allows for real-time translation: speak in English, and ChatGPT can translate your words into Spanish or Hindi with only a short delay. When the AI needs to think harder, it can hand off complex reasoning to a more powerful model, GPT-5.5, and keep chatting with you while it works.

The voice now includes human-like filler words — “um,” “er,” “like” — and responds faster when you cut in. You can also set it to listen silently until you say its name, similar to a wake word. OpenAI says audio clips are stored for 30 days to maintain context, but users are automatically opted out of training on voice data. The company also notes expanded safety filters around self-harm, violence, and other sensitive topics, acknowledging the risks of making AI feel too human.