OpenAI has launched GPT-Live, a new generation of voice models now powering ChatGPT Voice. According to OpenAI, the models use a "full-duplex" architecture, meaning they can listen and speak at the same time — closer to how two people actually talk over each other.
The company is shipping two versions. Per Sabrina Ortiz of The Deep View (via Techmeme), GPT-Live-1 powers ChatGPT Voice for Go, Plus, and Pro users, while GPT-Live-1 mini is the default for free accounts.
The pitch is a more human-feeling conversation. The Verge reports the new GPT-Live-1 is designed to interrupt you less and will wait for you to keep going if you pause mid-sentence, with OpenAI describing it as more like "talking to another person." Engadget notes the model will acknowledge you as you speak and slow down if you ask it to, and Cybernews observed it can even drop in a "mhm" like a real listener.
Under the hood, the voice model doesn't do all the heavy lifting itself. The Decoder and MarkTechPost report that complex questions get handed off to GPT-5.5 in the background, which improves the quality of answers, and that the system delegates search and reasoning to that larger model. Search Engine Journal notes GPT-Live also brings web search into ChatGPT Voice.
The launch fits a broader strategy: Axios reports OpenAI is betting that voice will become AI's primary interface. The Decoder adds that GPT-Live-1 is available now for paying ChatGPT users, with the mini version for free accounts.
Why it matters: smoother, interruptible voice chat could make talking to AI feel less like issuing commands and more like a natural conversation — a step toward voice replacing typing as the main way people use these tools.