OpenAI has announced GPT-Live, a new generation of voice models powering the latest ChatGPT Voice experience. Unlike previous voice systems that relied on turn-based interactions, GPT-Live introduces a full-duplex architecture, allowing ChatGPT to listen and speak simultaneously for more natural, real-time conversations.

The company said GPT-Live also works alongside GPT-5.5, which handles complex reasoning, web searches, and other advanced tasks in the background while conversations continue uninterrupted.

OpenAI explained that earlier versions of ChatGPT Voice relied on separate speech-to-text, language, and text-to-speech models, which could introduce delays and information loss.

Although Advanced Voice Mode unified audio processing into a single model, conversations remained turn-based, requiring users to finish speaking before ChatGPT responded.

GPT-Live replaces that approach with continuous audio processing, enabling the model to:

• Listen and speak simultaneously
• Decide several times per second whether to respond, continue listening, pause, interrupt, or use tools
• Maintain more natural conversational flow
• Support live translation
• Continue conversations while background tasks are processed

When users request web searches, deeper reasoning, or agentic tasks, GPT-Live delegates those requests to GPT-5.5 while continuing the voice conversation.

According to OpenAI, GPT-Live can manage multiple background tasks simultaneously and return results once they are complete. The company added that future frontier models will also integrate with GPT-Live as they become available.

OpenAI said more than 150 million people use ChatGPT Voice and Dictation every week.

The updated ChatGPT Voice experience includes:

• Full-duplex conversations with simultaneous listening and speaking
• Natural acknowledgements such as “mhmm” and “got it”
• Instant, Medium, and High reasoning modes
• Fewer interruptions when users pause during conversations
• Improved background noise handling
• Rich visual cards for weather, sports, and stocks
• Continued support for search, memory, images, and file uploads
• Remastered versions of ChatGPT’s nine built-in voices

According to OpenAI, GPT-Live-1 and GPT-Live-1 mini outperformed the previous Advanced Voice Mode in internal human evaluations covering five- to ten-minute conversations.

The company also reported improvements across several benchmarks, including:

• GPQA for expert-level scientific reasoning
• BrowseComp for agentic web search and difficult-to-find information
• τ³-Voice Telecom, an internal benchmark measuring realistic multi-turn telecom support conversations

OpenAI said GPT-Live introduces safeguards designed specifically for voice interactions.

Safety measures include:

• Real-time interventions while the model is speaking
• Safer response redirection when needed
• Ending conversations in higher-risk situations
• Expert-reviewed crisis helpline guidance for self-harm discussions
• Age-appropriate experiences for teenagers
• Parental controls with notifications in certain high-risk situations
• Protection against voice impersonation through predefined ChatGPT voices
• Ongoing monitoring for emotional reliance

The company said GPT-Live has also undergone expanded testing covering self-harm, psychosis, mania, violence, sexual content, and other voice-specific safety risks.

GPT-Live is rolling out globally across iOS, Android, and ChatGPT.com.

Availability includes:

• GPT-Live-1 as the default voice model for Go, Plus, and Pro users
• GPT-Live-1 mini as the default voice model for Free users

OpenAI said API support will be introduced in a future release.

At launch, GPT-Live does not support voice conversations with video or screen sharing. Users can continue using the legacy Standard Voice and Advanced Voice Mode for those features while OpenAI works to bring video and screen sharing support to GPT-Live in a future update.