Perplexity voice mode just got the one upgrade that makes voice assistants feel human: it listens while it speaks. On October 28, 2026, Perplexity upgraded its Mac app and web voice mode to run on OpenAI’s gpt-live-1 model, the same model that powers ChatGPT’s voice.
For creators and researchers who work by voice, the practical change is bigger than sound quality. You can interrupt, correct, or redirect the assistant mid-sentence, and you can run several Perplexity Computer sessions at once without touching the keyboard.
What actually changed
The headline capability of gpt-live-1 is full-duplex voice. Most voice assistants process input in discrete chunks: you talk, it talks, you wait. Full-duplex keeps the audio stream open, so the model keeps listening while it is still speaking, much like a conversation between people.
Perplexity pairs that conversational layer with real actions. In one voice session, Pro and Max subscribers can now:
- Work across multiple Computer sessions at once. Direct several agent tasks in parallel by speaking, instead of switching tabs and retyping context.
- Start new tasks mid-conversation. Kick off research or agent work without ending the session you are in.
- Create and manage Projects by voice. Set up and organize work without moving to the text interface.
Computer is Perplexity’s AI agent, which plans and carries out tasks on the user’s behalf. The voice upgrade turns it from a tool you dictate to into a collaborator you talk with while it works.
The evidence: model, availability, and the local-AI pairing
| Detail | What the source says |
|---|---|
| Announced | October 28, 2026, via Perplexity’s Computer account on X |
| Voice model | OpenAI’s gpt-live-1, the model behind ChatGPT’s voice |
| Key capability | Full-duplex voice: listens while speaking, so users can interrupt or redirect mid-response |
| Where available | Perplexity Mac app and web interface, from October 28 |
| Who gets it | Pro and Max subscribers only |
| Multi-session | Work across multiple Computer sessions at once in one voice session |
| Project control | Create and manage Projects by voice |
| gpt-live-1 history | Reached ChatGPT Voice users July 8, 2026; API version opened September 10, 2026 at $0.05 per minute for the voice layer, per coverage |
| Related Perplexity launch | October 20, 2026: local AI execution on supported Macs, for offline and privacy-focused work |
| What it is not | A new Perplexity-built voice model — Perplexity adopted OpenAI’s model rather than building its own |
Full-duplex means both directions of a conversation stay open, so either side can interrupt at any moment. Computer sessions are Perplexity’s agent task workspaces, which voice now lets you steer several at once.
A voice-first research session with Perplexity
1. Open voice on Mac or web — Pro or Max subscription required.
2. Start a Computer session by voice — Describe the task out loud.
3. Interrupt and redirect — Correct course mid-answer instead of waiting for a full stop.
4. Spin up parallel sessions — Start a second task while the first still runs.
5. File it in a Project — Create or manage the Project without touching the keyboard.
How to profit from it
Voice that interrupts changes how fast solo operators can work. Sell the workflow, not the feature:
- Offer voice research sprints. Run client research, competitor scans, and briefing docs in live voice sessions, and deliver the write-up the same day.
- Sell a Perplexity Computer setup. Configure Projects, sessions, and task templates for a client, then train them to drive it by voice in one session.
- Pair voice with local AI for sensitive work. Perplexity’s October 20 local execution runs models on the Mac itself. Pitch the split honestly: local for private data, cloud voice for heavy research.
- Build a dictation-to-deliverable service. Creators dictate briefs, outlines, and scripts; you run the voice sessions and return finished drafts.
- Charge for the interruption habit. The skill is knowing when to cut the model off and redirect it. That judgment is what clients pay for, and it is learned by doing, not by reading a changelog.
What to watch before you switch
First, this is a paid-tier feature. Free users do not get the new voice mode, and there is no announced date for wider access.
Second, Perplexity did not build this model. The voice layer depends on OpenAI’s gpt-live-1, so its quality, limits, and pricing direction are partly OpenAI’s decisions, not Perplexity’s.
Finally, voice work still needs a human at the end. Read the deliverable before it goes out.
Sources: Perplexity’s October 28 announcement as reported by 9to5Mac and CryptoBriefing, linked below. The announcement itself was posted on X on October 28, 2026; figures are reported from that coverage.
Frequently Asked Questions
What changed in Perplexity’s voice mode?
On October 28, 2026, Perplexity upgraded its Mac app and web voice mode to run on OpenAI’s gpt-live-1 model, adding full-duplex voice that listens while it speaks.
What is full-duplex voice?
A conversation mode where the audio stream stays open in both directions, so you can interrupt, correct, or redirect the assistant while it is still speaking, instead of waiting for it to finish.
Who gets the new voice mode?
Perplexity Pro and Max subscribers, on the Mac app and the web interface, from October 28, 2026.
Can I run several tasks by voice at once?
Yes. Subscribers can work across multiple Perplexity Computer sessions at once, start new tasks mid-conversation, and create and manage Projects in a single voice session.
Did Perplexity build its own voice model?
No. It adopted OpenAI’s gpt-live-1, the model behind ChatGPT’s voice, which first reached ChatGPT Voice users on July 8, 2026.
Sources: 9to5Mac — Perplexity upgrades Mac app with ChatGPT-powered voice · CryptoBriefing — Perplexity taps OpenAI’s gpt-live-1
Keep Reading
Want more tools that actually pay for themselves? Browse the AI Tools section for the latest launches, pricing breakdowns, and profit playbooks — and check the official source linked above before you buy or deploy anything.