ChatGPT Voice Mode Now Controls Your Computer

> **TL;DR:** OpenAI's ChatGPT desktop app now lets voice mode control your computer: it can open applications, click UI elements, and complete full tasks on Mac and Windows for all paid plans. The feature runs on GPT live, a new voice model family that listens and speaks simultaneously, and ChatGPT can delegate work to other AI agents while you keep talking.

Key Takeaways

- ChatGPT's desktop voice mode can now open apps, click UI elements, and finish whole tasks on both Mac and Windows. - The capability is included with every paid ChatGPT plan — not just a top-tier subscription. - It runs on GPT live, a new family of voice models that listen and speak at the same time, so you can interrupt mid-task. - ChatGPT can dispatch background work — checking email, fixing code, completing work-feature tasks — to other agents while the conversation continues. - Voice is shifting from a chat interface to the control layer for desktop AI agents.

OpenAI has turned the ChatGPT desktop app's voice mode into something far more consequential than a conversation feature: it can now operate your computer. Speaking to ChatGPT on Mac or Windows, users can have the assistant open applications, click UI elements, and carry complete tasks through to the finish — hands-free. The capability is available on every paid ChatGPT plan, and it runs on GPT live, a new family of voice models designed to listen and speak at the same time, so you can interrupt and redirect the AI mid-task the way you would a human colleague. Most striking of all: while you talk, ChatGPT can quietly dispatch other AI agents to handle work in the background, turning a spoken conversation into a managed, multi-agent workflow.

Here's what changed, why the underlying model matters, and where this fits in the broader race to make agents genuinely useful on the desktop.

Voice Mode Grows Hands

The headline change is computer control. Until now, voice interfaces to chatbots have mostly been an input-output swap: you speak instead of type, the model talks instead of writes. The updated ChatGPT desktop voice mode breaks that pattern. Ask it to do something on your machine, and it can open the relevant application, navigate the interface by clicking UI elements, and see the task through to completion — not just describe the steps, but perform them.

Two details make this more than a demo. First, it ships on both macOS and Windows, so the overwhelming majority of desktop users are covered rather than the usual Mac-first slice. Second, access is not gated behind a premium tier: any paid ChatGPT plan gets it.

The framing matters. Computer use — agents that perceive a screen and act on it with clicks and keystrokes — has been one of the defining research threads across the AI industry. What OpenAI has done here is change the steering mechanism. Instead of typing a goal into a text box and watching an agent grind through it, you talk to your computer while it works, correcting course out loud as things unfold. That's a different product, not just a different modality.


One Conversation, Many Agents

The second piece is arguably the bigger architectural story: ChatGPT can now delegate. While you're mid-conversation by voice, it can simultaneously send work off to other agents — checking your email, fixing code, or completing tasks inside ChatGPT's work features — and keep talking to you while that happens.

That's the orchestrator pattern arriving in a mainstream consumer product. Developers have been assembling exactly this kind of architecture by hand for some time — wiring up planner agents that fan work out to specialized workers — and the pattern dominates the agentic tooling we highlighted in our [trending GitHub repositories roundup](https://speka.info/blog/trending-github-repos-opencode-firecrawl-5-more), from terminal coding agents to scraping pipelines. Anthropic has been teaching the same ideas through official developer examples, as we covered in our look at the [Claude Cookbook](https://speka.info/blog/claude-cookbook-anthropics-official-recipe-hub-goes-viral). The difference is that those are patterns you build. This is a pattern you get by talking.

The practical upshot: a voice session stops being a single-threaded exchange. You can be discussing one problem while the assistant has already put background work in motion elsewhere — a fan-out model of personal computing where your role shifts from operator to manager.

GPT Live: Why Full-Duplex Voice Is the Real Enabler

Powering all of this is GPT live, OpenAI's new family of voice models. Their defining trait is simultaneity: they can listen and speak at the same time rather than taking rigid turns, which means you can cut in mid-sentence — or mid-task — and redirect the AI naturally, the way you'd interrupt a coworker heading down the wrong path.

It's tempting to file that under nicer conversations, but for an agent with control of your computer it's closer to a safety-critical feature. An assistant that clicks through your applications needs a real-time interrupt channel. If the only way to stop a misfiring action is to wait for the model to finish talking, complete its turn, and then process your objection, small mistakes get expensive. Full-duplex listening turns "no, wait, not that window" into an instruction the agent can act on immediately.

This also explains why OpenAI shipped computer control and the new voice family together. The two capabilities aren't independent features that happen to share a release; each one makes the other viable. Voice without agency is a novelty. Agency without live steering is a liability.


What It Means for the Desktop AI Race

A few observations worth drawing out from this release.

The desktop is back in play

For years the assumption was that AI assistants would live in the browser or the chat window. A voice agent that operates native applications makes the operating system itself the surface — and puts competitive pressure on Microsoft, Apple, and Google, whose platforms now host an increasingly capable third-party agent.

Accessibility gets a genuine upgrade

Hands-free computing has historically meant brittle command grammars and plain dictation. A voice agent that can carry out full tasks — not just transcribe — is a meaningful step for users who can't comfortably use a keyboard and mouse, and a plausible new default for anyone whose hands are simply busy.

The trust questions arrive with the convenience

As with any computer-use agent, day-to-day usefulness will hinge on scoping: what the assistant can see, which actions require confirmation, and how easily users can bound its reach. Those operational details will matter as much as raw capability, and they're the thing to watch as the feature settles into real workflows.

Availability and Open Questions

Here's what's confirmed: the new voice capabilities are live in the ChatGPT desktop app on both Mac and Windows for all paid ChatGPT plans, they're powered by the GPT live model family, and ChatGPT can delegate work to other agents mid-conversation.

Plenty remains open. Access is specified for paid plans, and whether any of this reaches the free tier is unconfirmed; there's also no verified word yet on developer API access to the GPT live models. We'll update our coverage as those details firm up.

The release lands amid an unrelenting stream of launches — our recent roundup of [GLM 5.2, OpenHuman, and free Claude Code access](https://speka.info/blog/glm-5-2-openhuman-free-claude-code-new-ai-tools) is a reminder that voice-driven agents are only one front. For ongoing coverage of model launches and feature rollouts like this one, follow our [LLM Launches & Updates hub](https://speka.info/llm-updates/).

The through-line is hard to miss. Chat interfaces made language models useful; agent frameworks made them productive; voice with computer control is a bid to make them ambient. If GPT live delivers natural interruption at scale, talking to your computer may finally stop feeling like a party trick — and start feeling like delegation.

Frequently Asked Questions

What can ChatGPT voice mode do on desktop now?

Beyond conversation, it can open applications, click UI elements, and carry out complete tasks on your computer, all directed by voice.

Which platforms and plans get the new voice capabilities?

The feature is available in the ChatGPT desktop app on both Mac and Windows, for any paid ChatGPT plan.

What is GPT live?

GPT live is OpenAI's new family of voice models that can listen and speak at the same time, so users can interrupt and redirect the AI mid-task in a natural real-time conversation.

Can ChatGPT work on other things while you're talking to it?

Yes. During a voice conversation it can delegate tasks to other AI agents — for example checking email, fixing code, or completing tasks inside ChatGPT's work features — while the conversation continues.

Is the feature available on the free ChatGPT plan?

Availability is specified for paid ChatGPT plans. Whether it will reach the free tier is unconfirmed.