OpenAI has launched GPT-Live, a new voice model that allows users to have more natural conversations with ChatGPT by allowing AI to listen and speak simultaneously.
The company said the model is rolling out globally starting today on iOS, Android, and ChatGPT.com, with GPT-Live-1 becoming the default voice model for Go, Plus, and Pro users, while GPT-Live-1 mini will power the Free tier.
Unlike previous versions of ChatGPT Voice that processed conversations in turns, GPT-Live is built on a full-duplex architecture that continuously processes audio as it generates speech. This allows users to interrupt the model, pause without being cut off, and have faster back-and-forth conversations.
For queries that require web search or deeper reasoning, GPT-Live hands the task to OpenAI’s latest frontier model while continuing the conversation. At launch, GPT-5.5 powers these background tasks, and OpenAI said future frontier models will automatically replace it as they become available.
“We’re launching GPT-Live, a new generation of voice models that make talking with AI feel much more like having a real conversation,” OpenAI said.
The company said more than 150 million people use ChatGPT’s voice and dictation features every week. GPT-Live adds support for visual response cards for topics such as weather, stocks, and sports during voice conversations, while continuing to support search, memory, images, and file uploads. Users can also choose different reasoning levels, including Instant, Medium, and High, depending on the complexity of the task.
OpenAI said GPT-Live separates conversation handling from reasoning tasks. While GPT-Live manages the interaction, GPT-5.5 performs search and reasoning in the background, allowing the conversation to continue without waiting for those tasks to finish.
According to OpenAI’s internal evaluations, GPT-Live-1 and GPT-Live-1 mini were preferred over Advanced Voice Mode in conversations measuring conversational flow, interruptions, and overall interaction quality. The company also reported improvements on benchmarks including GPQA for scientific reasoning, BrowseComp for web search, and τ³-Voice Telecom for multi-turn customer support tasks.
OpenAI said it expanded safety testing for GPT-Live with audio-native evaluations covering areas such as self-harm, emotional reliance, violence, and sexual content.
The company added safeguards that can intervene during live conversations by steering responses, surfacing safety resources, or ending conversations in higher-risk situations. It also introduced protections for teen users through parental controls and age-appropriate behaviour.
“GPT-Live was designed to be safe by default,” the company said.
At launch, GPT-Live does not support voice conversations with video or screen sharing, though OpenAI said those capabilities are planned for a future release. Users who need those features can continue using the legacy versions of ChatGPT Voice, including Standard Voice Mode and Advanced Voice Mode.
GPT-Live is OpenAI’s latest effort to make voice interaction a primary interface for AI systems by combining continuous conversation with its latest reasoning models in the background.
Your reaction
Discussion
What's your take on this story?
No comments yet — be the first to share your view.