OpenAI Introduces GPT-Live Voice Model for Natural Conversations
The shift to full-duplex architecture addresses the unnatural pauses and interruptions that plagued previous voice modes, moving closer to human-like conversation.
Reporting from 1 source: GIGAZINE.
OpenAI announced GPT-Live on July 8, 2026, a new voice model for ChatGPT that uses full-duplex architecture to allow simultaneous speaking and listening. The model supports backchanneling, interruptions, and natural pauses, moving beyond the turn-based limitations of previous voice modes. It delegates complex tasks to GPT-5.5 in the background and offers three reasoning levels. Nine remastered voices are included. Human evaluations showed strong preference for GPT-Live over advanced voice mode.
OpenAI's GPT-Live replaces the turn-based structure of ChatGPT's advanced voice mode with a full-duplex system that processes input and output simultaneously. The model decides many times per second whether to speak, listen, pause, interrupt, or call tools, enabling interjections like "uh-huh" and quiet waiting while the user thinks. For heavier tasks like web search or deep reasoning, GPT-Live delegates to GPT-5.5 in the background, preserving conversation flow. Users can select from three reasoning levels-Instant, Medium, and High-with corresponding GPT-5.5 configurations. The nine available voices have been remastered to better focus on the user's voice amid background noise. OpenAI's internal evaluations found GPT-Live strongly preferred over the previous advanced voice mode across measures of turn-taking, naturalness, and overall conversational comfort.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.