Perplexity's voice mode now runs on OpenAI's gpt-live-1, which listens while it answers

When you talk to an AI, you may know the feeling: You have to wait until it finishes before you can say anything. A new model in Perplexity is meant to fix that. The company announced on October 28, 2026, that its voice mode now runs on OpenAI's gpt-live-1 model. One of the sources mentions October 29, probably because of the time difference.

Listens while it answers

gpt-live-1 is what is known as full-duplex. That means the model keeps listening while it is answering. Several times a second, it decides whether to talk or listen. It resembles an ordinary conversation, where people can interrupt each other.

If you ask a hard question that requires a web search or deeper reasoning, gpt-live-1 hands the task off to a backend model. That is a different model that works in the background while the conversation continues. The sources do not say which backend model Perplexity uses or which languages the voice mode supports.

Control multiple tasks by voice

With the new voice mode, users can manage several Computer sessions at the same time, start new tasks, and create and manage projects in a single voice session.

The upgrade is available in Perplexity's Mac app and on the web, for Pro and Max subscribers. Perplexity has not said whether free users will get access. Pro costs $17 a month with annual billing, and Max costs $167 a month with annual billing.

Note that the requirement for macOS 15 Sequoia or later applies to the Personal Computer agent on macOS, that is, the Mac part. It has not been stated as a requirement for voice mode on the web.

From October 20, 2026, Computer also runs fully locally on Mac for Pro, Max and Enterprise subscribers on supported Apple silicon Macs.

From ChatGPT to Perplexity

OpenAI released gpt-live-1 to ChatGPT Voice users on July 8, 2026, and opened an API version on September 10, 2026. An API is a technical access point that lets other companies use the model in their own products. Developers pay $0.05 per minute for the voice layer.

According to Crypto Briefing, this could create financial tension for Perplexity. The company charges fixed monthly prices but pays per minute for the API, so costs could rise the more users talk.