How do I switch from ChatGPT Advanced Voice to GPT-Live?

GPT-Live is rolling out sequentially to ChatGPT users worldwide starting July 8, 2026, so you may not need to switch manually — it becomes the default voice mode when your account is upgraded. On paid plans (Go, Plus, Pro) you get GPT-Live-1; on Free you get GPT-Live-1 mini. If you don't see GPT-Live yet: update the ChatGPT app (iOS/Android/web), sign out and back in, or wait 1-2 weeks for staged rollout to reach your region. In-app: tap the voice icon and look for 'GPT-Live' badge on the mode.

What breaks when I switch from Advanced Voice to GPT-Live?

Very little user-facing. GPT-Live handles voice conversation in real time and delegates heavy reasoning to GPT-5.5 in the background — the same underlying reasoning model Advanced Voice uses. Custom Instructions, Memory, and voice selection (Ember, Vale, Cove, etc.) carry over. What changes: full-duplex conversation (you can interrupt, it backchannels), three reasoning levels (Instant/Medium/High), rich visual cards for weather/stocks/sports, and no more turn-taking pauses. If you built third-party workflows on the Realtime API, those keep working — Realtime API is separate and still available.

Should I switch back to Advanced Voice if I prefer turn-based?

For now, no toggle exists — GPT-Live becomes the default when it reaches your account. OpenAI has not committed to keeping Advanced Voice as an opt-in. If you rely on turn-based UX (accessibility reasons, screen readers, hands-free workflows tuned to fixed pauses), file feedback in-app. The safer path if you built accessibility around Advanced Voice: use the Realtime API in third-party clients (which remains turn-based via the API's response.done events) instead of the ChatGPT app.

Can I use GPT-Live in my own app or via API?

Not yet, as of July 12, 2026. OpenAI said API access is planned but did not give a launch date at the July 8, 2026 announcement. The Realtime API (turn-based) is still available and priced at roughly $0.06/min input + $0.24/min output. If you need full-duplex voice in your own product this quarter, use ElevenLabs Conversational AI, Retell, Vapi, or Bland. Watch OpenAI's DevDay in October 2026 for likely GPT-Live API announcements.

Quick Answer

How to Switch from ChatGPT Advanced Voice to GPT-Live (July 2026)

Published: July 12, 2026

The 60-Second Version

What changed: On July 8, 2026, OpenAI announced GPT-Live — a full-duplex voice model that replaces Advanced Voice as ChatGPT’s default voice mode.
How you get it: Automatic staged rollout to ChatGPT users worldwide starting July 8-9, 2026. No opt-in required.
What’s better: Interruption handling, natural backchannels, three reasoning levels, visual cards during voice, sub-500ms latency.
What’s the same: Voice selection, Custom Instructions, Memory, availability on iOS/Android/web.
What’s not available yet: API access. Only ChatGPT app for now.

Step-by-Step Migration

1. Update the ChatGPT app

iOS: App Store → ChatGPT → Update. Minimum version 1.2026.7 or later.
Android: Play Store → ChatGPT → Update. Minimum version 1.2026.7 or later.
Web: chatgpt.com in Chrome/Safari/Firefox — hard refresh (Cmd/Ctrl-Shift-R). No install needed.

2. Sign out and back in

Some users report GPT-Live doesn’t appear until a session refresh. Sign out, sign in, restart the app.

3. Look for the badge

Open a chat, tap the voice icon (waveform, bottom right on mobile). If it says “GPT-Live” at the top of the voice UI — you’re on the new model. If it says “Advanced Voice” — you’re still on the old one; wait 1-2 weeks for rollout.

4. Pick a reasoning level

New in GPT-Live: three reasoning levels for background thinking.

Level	Latency	Use for
Instant	~250ms	Chat, quick questions, dictation
Medium	~600ms	Coding help, planning, analysis
High	~1500ms	Research, deep reasoning, complex tasks

Toggle inside the voice UI. Instant is default.

5. Try the killer features

Interrupt it: Talk while it’s talking. It will stop mid-sentence and listen.
Let it interrupt you (backchannel): Take a natural pause; you’ll hear “mm-hmm” or “right” — it’s tracking.
Ask it to look something up: “What’s the weather in Tallinn?” GPT-Live delegates to GPT-5.5 in the background and shows a weather card.

What Actually Changed Under the Hood

Model: GPT-Live-1 (paid) or GPT-Live-1 mini (free) — a voice-specialized model, not the full frontier.
Reasoning delegation: When you ask for something complex, GPT-Live passes the task to GPT-5.5 running in the background and keeps talking to you while it processes.
Full-duplex architecture: Listen and speak simultaneously (previously: turn-based).
Safety layer: Voice-native safety evals, real-time monitoring, uses predefined ChatGPT voices only (no cloning of arbitrary voices).
Data retention: Audio clips from Live and Advanced Voice conversations are kept for 30 days with the chat transcript. Not used for training unless you opt in.

What Did Not Change

Voices: Ember, Vale, Cove, Sol, Breeze, etc. all carry over.
Memory / Custom Instructions: unchanged.
Free tier availability: Free users still get voice (now GPT-Live-1 mini).
Realtime API: still available separately, still turn-based, still priced ~$0.06 in / $0.24 out per minute.
Third-party voice AI: ElevenLabs, Retell, Vapi, Bland — none affected.

If You Don’t See GPT-Live Yet

Rollout is sequential and regional. Reasons you might not have it:

Staged rollout not reached your region yet. Wait 1-2 weeks.
Old app version. Update from the app store.
Session cache. Sign out, sign in.
Enterprise / Team account with delayed rollout. Enterprise admins may hold GPT-Live back for review. Ask your admin.
You’re already on it and didn’t notice. Look for the “GPT-Live” badge in the voice UI header.

What to Do If You Prefer Advanced Voice

As of July 12, 2026 there is no in-app toggle to switch back. OpenAI’s announcement did not promise Advanced Voice as an opt-in. If your workflow depends on turn-based UX:

File feedback in the ChatGPT app (👍/👎 → “Report bug or suggestion”)
Use the Realtime API in a third-party client (Reasoning ChatGPT alternatives, OpenAI’s own API playground) — Realtime remains turn-based
Consider ElevenLabs Conversational AI with a turn-based configuration

When GPT-Live API Will Ship (Speculation)

Announced: not yet, as of July 12, 2026
Rumored window: OpenAI DevDay is typically October. GPT-Live API is the highest-probability October 2026 announcement.
Expected pricing: probably ~$0.05-$0.10/min in, ~$0.20-$0.40/min out for GPT-Live-1, cheaper for GPT-Live-1 mini
Realistic ETA for production usage: Q4 2026 preview, Q1 2027 GA

If you’re building now, do not wait — build on Realtime API or a third-party (Retell/ElevenLabs/Vapi) and plan a Q1 2027 swap.

Sources

OpenAI GPT-Live announcement coverage: Macrumors: OpenAI GPT-Live voice model
Feature deep-dive: VentureBeat: OpenAI launches GPT-Live, a full-duplex voice upgrade
Developer perspective: Full-Duplex Voice AI Architecture