Voice is becoming the front end for AI agents. GPT-Live is the version of that you can use today, inside ChatGPT, without writing any code. It listens while it talks, so you can interrupt and change direction mid-sentence. Here's the setup, what to test first, and where it still breaks.
Check which Voice you're on
Open Settings → Voice. What you see depends on your plan, region and app version.
| Option | What it is | Use it for |
|---|---|---|
| Live | GPT-Live-1, or GPT-Live-1 mini on Free and Go. Listens and speaks at the same time | Natural conversation, thinking out loud, files and Projects |
| Advanced | The previous real-time Voice | Live video or screen sharing on iOS and Android |
| Standard | Transcribes your speech, then answers turn by turn | When you want strict turn-taking |
Daily Live limits (rolling 24 hours): Go gets 3 hours of GPT-Live-1 mini, Plus gets 3 hours of GPT-Live-1, Pro ($100) gets 15 hours and Pro ($200) is unlimited. Free has limited mini access.
Set it up in two minutes
- Select Live: Settings → Voice → Live.
- Pick a voice. There are nine. Switching voices mid-call starts a new call in the same chat.
- Set your language: Settings → Voice → Language. Pick the one you speak most for better recognition.
- Turn on Background conversations to keep talking with your phone locked or in another app.
- On iPhone: during a call, open Control Center → Mic Mode → Voice Isolation. This cuts false interruptions in noisy places.
- For hard questions, choose the model and reasoning effort with the same controls as text chat. Voice can hand off to GPT-5.6 or GPT-6 Astra when it needs to search or reason.
Then tap the Voice icon in the message bar and start talking.
Six things to try
1. Think out loud, then read the result. Tell it to wait first, so it doesn't jump in during your pauses.
Wait until I say "go" before responding. I'm going to talk through a messy idea.
[talk]
Go. Turn that into a one-page plan with open questions at the end.
Read the plan as text afterward. Voice is fast for input. A screen is better for review.
2. Rehearse a hard conversation. Interviews, pitches, salary talks. Interrupting it is the whole point.
You're a skeptical hiring manager for [role]. Ask me one question at a time and push back on vague answers. After five questions, tell me my weakest answer and why.
3. Talk through a document. Attach a file with the + button during the call.
Walk me through the three decisions this document is asking me to make. Stop after each one so I can react.
4. Use it inside a Project. Voice can reference the project's recent chats, sources and instructions, which makes a quick status check easy.
Based on this project's recent chats, what did I leave unfinished, and what should I do first today?
5. Practice a language. Ask for corrections on a schedule so the conversation keeps flowing.
Let's talk in Spanish about my weekend. Don't correct me mid-sentence. Every few turns, give me my two biggest mistakes.
6. Hands-busy planning. Walking or cooking. Keep it to lists you'll check later.
Help me plan tomorrow. Ask me what's due, then read back my top three priorities.
What's good
- Interruptions feel natural. You can cut in, add a detail or change direction without waiting for it to finish.
- It stays in one chat. Replies stream as text while it speaks, and you can type or add images without leaving Voice.
- It has more context than older Voice. Web search, memory, files and Projects all work in Live.
Where it falls short
- No video or screen sharing. Use Advanced for that.
- No connected apps or plugins yet. It can't act in your tools from a Live call.
- One speaker at a time. It can respond to people talking to each other nearby.
- Long pauses can still trigger a reply, even after "wait until I say go."
- Transcripts aren't verbatim. Don't treat them as a record of what was said.
- Audio clips are kept for 30 days with the chat transcript. They're only used for training if you opt in under Settings → Data Controls.
- It can be wrong out loud. Verify dates, numbers and anything location-specific.
Your first 15 minutes
- Settings → Voice → Live, set your language and voice
- Turn on Voice Isolation (iPhone) or use headphones
- Try #1: a 3-minute brain dump, then read the plan it writes
- Try #3 or #4 on something real you're working on
- Interrupt it on purpose twice, and notice whether it follows the change
- Decide on one job: brainstorming, rehearsal, document Q&A or daily check-in. Use Voice only for that job for a week.
Building a voice agent? The same model is in the API as gpt-live-1 at $0.05 per minute, plus backend usage. Read how the voice and backend split works, then the GPT-Live quickstart.
Go deeper: Voice Is Becoming the Front End for AI Agents · ChatGPT, End to End · Follow-Up Prompts That Fix Weak AI Answers · ChatGPT Voice help