I tested ChatGPT's Live Voice upgrade, and it almost felt human - how to try it
TL;DR
OpenAI’s new GPT-Live voice models can listen, speak, and do online research at the same time, making conversations feel less like turn-based chatbot exchanges. In testing, ChatGPT sounded almost human in moments: shorter pauses, smoother turn-taking, and small listening cues made the interaction feel more natural. Users can try it through ChatGPT Voice on web or mobile. Paid tiers get the stronger GPT-Live-1 model, while free users are expected to get a lighter Mini version.
Nauti's Take
This is a meaningful step, but also a very OpenAI kind of moment: impressive product magic wrapped in plenty of PR shine. A more natural voice does not automatically mean better reasoning.
The real test is whether ChatGPT stays accurate during interrupted, messy, multi-step tasks. For everyday use and research it could be powerful, as long as users do not confuse a human-like voice with truth.
Briefingshow
This moves voice AI from a demo feature toward a real interface: people can talk while cooking, driving, planning, or researching without carefully waiting for turns. The bigger question is reliability, not just naturalness. When an assistant researches while speaking, usefulness rises, but so does the risk that a confident voice makes weak answers feel more trustworthy than they are.
Sources