What Is GPT-Live?

OpenAI introduced GPT-Live, a new generation of voice models that powers ChatGPT Voice. The key capability: GPT-Live can listen and speak at the same time.

Try Apidog today

Previous ChatGPT voice experiences, including Advanced Voice Mode, were turn-based: you talked, it waited, then it answered. GPT-Live is full-duplex. It processes incoming audio while generating output, then decides many times per second whether to speak, keep listening, pause, interrupt, or call a tool. That enables more natural behaviors like backchanneling with “mhmm,” waiting while you think, and recovering when you interrupt it mid-sentence.

GPT-Live at a glance

	Detail
Announced	July 8, 2026
What it is	Full-duplex voice model family powering ChatGPT Voice
Variants	GPT-Live-1, GPT-Live-1 mini, GPT-Live-1 Medium, GPT-Live-1 High
Backend	GPT-5.5 Instant (1 and mini); GPT-5.5 Thinking (Medium and High)
Default for paid (Go/Plus/Pro)	GPT-Live-1
Default for Free tier	GPT-Live-1 mini
Platforms	iOS, Android, chatgpt.com (rolling out globally)
Video / screen sharing	Not at launch; “working to introduce these capabilities soon”
API	Not available yet; OpenAI plans to bring it to the API “soon”

How the GPT-Live variants map to backend models

GPT-Live ships as a model family. OpenAI describes the backend split clearly:

“GPT-Live-1 (instant) and GPT-Live-1 mini use the GPT-5.5 Instant model in the background, while GPT-Live-1 Medium and GPT-Live-1 High use the GPT-5.5 Thinking model with medium and high reasoning effort.”

In other words, GPT-Live is the voice interaction layer, while GPT-5.5 handles the heavier reasoning behind it.

Use the variants this way:

• GPT-Live-1: default daily voice model for Go, Plus, and Pro subscribers.
• GPT-Live-1 mini: default Free-tier voice model with the same full-duplex interaction style.
• GPT-Live-1 Medium / High: slower but deeper reasoning, backed by GPT-5.5 Thinking at medium or high effort.

The architecture pattern: full-duplex plus delegation

GPT-Live’s main architectural idea is not only full-duplex audio. It also delegates work.

When a question needs search, deeper reasoning, or more agentic behavior, GPT-Live can hand that task to another model such as GPT-5.5, then fold the result back into the voice conversation when ready.

That matters because voice agents have a common tradeoff:

• Realtime voice models need low latency.
• Deep reasoning models are usually slower.
• Users still expect the conversation to continue naturally.

GPT-Live splits those responsibilities:

1. The voice model keeps the conversation active.
2. It listens, speaks, pauses, and backchannels in realtime.
3. A stronger model works on harder tasks in the background.
4. The final result is returned into the conversation when available.

What GPT-Live can do at launch

At launch, GPT-Live supports:

• Web search delegated into the conversation flow
• Live translation between languages
• Visual cards for weather, stocks, and sports
• Image and file uploads during a voice conversation
• Memory, consistent with the rest of ChatGPT

The key limitation: GPT-Live does not support voice with video or screen sharing at launch.

Advanced Voice Mode still keeps video and screen sharing for eligible subscribers, and both Standard Voice Mode and Advanced Voice Mode remain available. If your workflow depends on pointing your camera at objects, screens, or documents, review GPT-Live vs Advanced Voice Mode before switching.

Benchmarks OpenAI cites

OpenAI compares GPT-Live mainly against Advanced Voice Mode, not competing products.

The cited evaluations include:

• GPQA: GPT-Live-1 “substantially outperforms” Advanced Voice Mode on expert-level scientific reasoning across biology, chemistry, and physics.

• BrowseComp: OpenAI reports “strong gains” on agentic web search and hard-to-locate information.

• τ³-Voice Telecom: an internal variant focused on realistic multi-turn telecom support tasks.

OpenAI also says users “strongly preferred” GPT-Live in 5-to-10-minute head-to-head conversations.

For developers, treat these claims as directional. OpenAI has not published detailed scores, latency numbers, or pricing yet. Independent comparisons will become more useful once the product is broadly available and testable. See the related GPT-Live vs Gemini Live comparison path for that context.

What developers can build today

There is no GPT-Live API today.

OpenAI’s current position is:

“We also plan to bring them to the API soon, and developers and enterprises can sign up to be notified using this form.”

If you are building voice agents now, the available path is the Realtime API, including:

• gpt-realtime-2.1 model family
• WebSocket transport
• WebRTC transport
• SIP phone calling
• MCP server support

For implementation details, start with Is there a GPT-Live API? What developers can use today. If you are comparing naming and architecture, read GPT-Live vs GPT-Realtime.

When prototyping against the Realtime API, Apidog can help with the non-model parts of the workflow:

• Testing WebSocket sessions
• Inspecting streamed events
• Mocking voice backend responses
• Keeping client development moving without requiring a live model connection for every test

How to try GPT-Live in ChatGPT

GPT-Live is rolling out globally across iOS, Android, and chatgpt.com, but the rollout is staged.

Use this checklist:

1. Update the ChatGPT app on iOS or Android, or open chatgpt.com.
2. Open Voice.
3. Check which model responds.
4. If you are on a paid plan, look for GPT-Live-1 as the default.
5. If you are on the Free tier, look for GPT-Live-1 mini.
6. If nothing has changed, the rollout likely has not reached your account yet.

For setup steps, variant switching, and the CarPlay path for all plans on iOS 26.4+, see how to use GPT-Live.

Why this launch matters for voice agents

Voice has been one of the most impressive AI demos, but one of the hardest interfaces to make production-ready.

Turn-based voice makes assistants feel like walkie-talkies:

1. User speaks.
2. Assistant waits.
3. Assistant responds.
4. User waits.

GPT-Live changes that interaction model by combining:

• Full-duplex listening and speaking
• Realtime interruption handling
• Natural backchanneling
• Delegation to stronger models for harder tasks

That creates a more realistic baseline for future voice agents. The developer API is not available yet, but OpenAI says it is coming “soon.” Until then, the practical implementation path is still the Realtime API.

If you are building a realtime voice system now, use the current Realtime API stack and test the transport layer carefully. WebSocket and streamed-event debugging are where many voice-agent bugs show up first.

FAQ

What is GPT-Live in one sentence?

GPT-Live is OpenAI’s full-duplex voice model family for ChatGPT Voice: it listens while speaking, backchannels naturally, and can delegate harder questions to GPT-5.5 in the background.

Is GPT-Live free?

Free-tier users get GPT-Live-1 mini as their default voice model. Paid tiers, including Go, Plus, and Pro, get GPT-Live-1.

Does GPT-Live support video or screen sharing?

Not at launch. Advanced Voice Mode still supports video and screen sharing for eligible subscribers. OpenAI says it is working to bring those capabilities to GPT-Live soon.

Is there a GPT-Live API?

Not yet. OpenAI plans to bring GPT-Live to the API “soon.” Developers can build voice agents today using the Realtime API.

What model powers GPT-Live?

GPT-Live-1 and GPT-Live-1 mini run on GPT-5.5 Instant. GPT-Live-1 Medium and GPT-Live-1 High use GPT-5.5 Thinking at medium and high reasoning effort.