OpenAI has launched GPT-Live, a new generation of voice models for ChatGPT that is designed to make AI conversations feel less like turn-taking and more like a real spoken exchange.
The headline change is full-duplex voice. GPT-Live can listen and speak at the same time, so it can acknowledge what a user is saying, handle interruptions, wait through pauses, and decide in the moment whether to respond, keep listening, pause, or invoke a tool.
That matters because voice AI has often felt unnatural for one simple reason: the model waits for a clean turn boundary. GPT-Live is OpenAI's attempt to move beyond that rigid back-and-forth.
Official OpenAI video
OpenAI's official GPT-Live video shows the new ChatGPT Voice experience as a more fluid conversation layer, with the assistant responding while the user is still shaping the task.
What GPT-Live changes
OpenAI says GPT-Live is built for continuous interaction rather than separate voice turns. Instead of waiting for a user to finish speaking, it continuously processes input while generating output.
In practice, that should make ChatGPT Voice better at:
- short acknowledgements like "mhmm" or "got it"
- quick back-and-forth conversation
- interruptions and mid-sentence course changes
- waiting while a user thinks
- filtering out background noise
- live translation
- showing visual cards for supported answers such as weather, sports, stocks, and maps
The other important design choice is delegation. GPT-Live handles the live conversation, but when a request needs web search, deeper reasoning, or more complex work, it can hand that task to a frontier model behind the scenes and bring the answer back into the voice conversation.
At launch, OpenAI says GPT-Live uses GPT-5.5 in the background. The company says it plans to update the delegated model as newer frontier models ship.
Availability
GPT-Live is rolling out globally in ChatGPT Voice across iOS, Android, and ChatGPT.com.
The model split is simple:
- GPT-Live-1 becomes the default ChatGPT Voice model for paid consumer users
- GPT-Live-1 mini becomes the default Voice model for Free users
OpenAI says the feature is rolling out across consumer plans, but availability can vary by region, app version, and account. The API is not generally available yet; OpenAI is collecting signups for developers and enterprises that want access to GPT-Live-1 in the API.
The video caveat
This is the part to be careful about: GPT-Live is a voice launch, not a live video launch.
OpenAI's own launch notes say GPT-Live does not support voice with video or screen sharing in ChatGPT at launch. Those capabilities remain in legacy Advanced Voice Mode for eligible users, while OpenAI says it is working to bring video and screen sharing to GPT-Live later.
So the near-term story is not "ChatGPT can now watch your screen in GPT-Live." The accurate story is that ChatGPT Voice is becoming more natural and more capable, while video and screen sharing are still coming.
Why this matters
GPT-Live is important because voice is becoming one of the main interaction layers for AI.
Typing works well for precise prompts. Voice works better when the user is moving, thinking out loud, driving, cooking, practicing a language, or trying to get help without stopping the task. But voice only feels useful when the interaction is fast, forgiving, and interruptible.
That is why full-duplex matters. If GPT-Live works well in real use, it could make ChatGPT feel less like a chatbot with audio and more like an assistant that can stay present in a natural conversation.
For OpenAI, it also sets up a larger product direction. A smoother voice layer can become the front door for deeper search, reasoning, translation, memory, and eventually more agentic work.
Safety and limits
OpenAI published a GPT-Live system card alongside the launch. The company says the models add voice-specific safety work, including checks that can happen during a live conversation and safeguards for areas such as self-harm, emotional reliance, sexual content, scams, manipulation, and impersonation.
The system card also says GPT-Live-1 and GPT-Live-1 mini are designed for predefined ChatGPT voices, not for imitating real people's voices.
The practical limits are also worth noting. GPT-Live is not initially available in ChatGPT Business, Enterprise, or Edu workspaces. It is also not initially available in Temporary Chats, the ChatGPT desktop app, Work, Codex, custom GPTs, video, screen sharing, connected apps, or plugins.
Our take
GPT-Live is a meaningful product update because it improves the part of AI that users feel immediately: the rhythm of conversation.
The launch is not about a bigger benchmark number. It is about whether ChatGPT can listen better, interrupt less awkwardly, respond faster, and keep a natural flow while still reaching for stronger models when the user asks something harder.
If OpenAI can bring this same interaction layer to video, screen sharing, Work, Codex, and API-based agents, GPT-Live could become more than a better Voice Mode. It could become the real-time interface for a much larger assistant platform.
For now, treat GPT-Live as a strong consumer voice upgrade with a clear roadmap signal: OpenAI wants ChatGPT to feel less like a text box and more like something you can talk to naturally.