GPT-Live brings natural voice and continuous dialogue with GPT-5.5 | Keryc
OpenAI announces GPT-Live, a new generation of voice models designed to make talking to an AI feel like talking to another person. Have you ever had an assistant's voice interrupt you at the worst moment or go quiet just when you wanted to keep going? GPT-Live tries to fix that with continuous interaction and smarter responses.
What is GPT-Live
GPT-Live is a voice model with a full-duplex architecture: it can listen and speak at the same time. That means it doesn't operate in rigid turns; it can show it's following you with small cues like "mhmm", interrupt you at the right moment, or stay quiet if you need time to think.
Also, when a question requires web search or deep reasoning, GPT-Live delegates that work to a frontier model in the background and brings the answer when it's ready. At launch, that background intelligence is GPT-5.5.
At first, two versions are released: GPT-Live-1 and GPT-Live-1 mini. They're available today for ChatGPT users on iOS, Android, and ChatGPT.com, and will soon arrive in the API for developers and businesses.
How it changes the conversation
GPT-Live introduces two key changes: continuous interaction and delegation for deep work.
Continuous interaction: instead of processing separate turns, the model processes input and output at the same time and decides several times per second whether to speak, listen, pause, or invoke a tool. Result? Smoother conversations, better timing, and features like live translation.
Delegation for deep work: GPT-Live handles the conversational part and sends complex tasks to more powerful models. That way it keeps the chat going while searching or doing long reasoning in the background.
In practice, this translates into features you already know but more natural: you can interrupt to correct it, ask it to wait while you think, or choose how much effort you want in the response (Instant, Medium, High).
Everyday examples: asking for directions during a trip, practicing a language while on the bus, dictating ideas for a document without losing the thread. The voice now complements visual actions: while you speak, ChatGPT can show visual cards with weather, actions, sports, and more.
Quick evaluations and performance
OpenAI ran human evaluations to measure preference, conversational flow, number of interruptions and naturalness. Results indicate that GPT-Live-1 and GPT-Live-1 mini are preferred over Advanced Voice Mode in 5–10 minute conversations.
Some cited benchmarks:
GPQA: better performance in scientific reasoning in biology, chemistry and physics.
BrowseComp: gains in web-search agent tasks and finding hard-to-locate information.
τ3-Voice Telecom: improved performance on multidimensional phone support tasks.
These numbers aren't magic; they show that the combination of continuous dialogue and access to frontier models improves both utility and the feeling of naturalness.
Safety designed for voice
Talking in real time requires different measures than text. GPT-Live includes new safety tests that cover audio evaluations and synthetic checks for areas like self-harm, psychosis, emotional dependence, violence and sexual content.
The system can intervene while speaking: redirect the response, show help messages, or end the conversation in higher-risk cases.
There are also protections for teens, crisis support flows and safeguards to prevent voice impersonation. OpenAI will continue monitoring real-world use to adjust wellbeing measures and reduce risks of emotional dependence.
Availability and limitations
Launch: global rollout on iOS, Android and ChatGPT.com. GPT-Live-1 will be default for Go, Plus and Pro users; GPT-Live-1 mini for Free users.
Languages: optimized for the most popular languages, but in some languages it may sound with a non-native accent or have gaps in fluency. They're working on improvements.
Limitations: at launch it doesn't support voice with video or screen sharing. You can still use previous ChatGPT Voice versions (Standard and Advanced) where those features are available.
Why this matters today?
Because the way we interact with AI stops being just asking for things and waiting for answers. GPT-Live aims to make conversation continuous, useful and less frustrating. It's not just a cosmetic upgrade: it changes how AI accompanies you in long tasks, in real-time support and in situations where tone and interruption matter.
Is it going to replace a human for all voice tasks? No. Does it make AI more natural and capable for everyday use? Yes. And that opens practical and ethical questions we'll figure out as it's used.
