Definition

Full-duplex voice AI processes audio input and generates speech output simultaneously, unlike cascaded STT→LLM→TTS pipelines or turn-based voice modes that wait for the user to finish speaking before responding.

Key Capabilities

  • Natural interruptions and back-and-forth without rigid turn-taking
  • Short acknowledgment cues (“mhmm”, “yeah”) while listening
  • Live translation while user continues speaking
  • Background task delegation while maintaining conversation flow

Paradigm Shift

openai gpt-live (July 2026) replaces Advanced Voice Mode’s turn-based architecture for 150M+ weekly ChatGPT voice users, representing a shift from sequential to continuous voice interaction.

Sources

Desktop Agentic Coding (2026-07-23)

gpt-live powers ChatGPT desktop voice for codex/chatgpt-work (2026-07-24-openai-gpt-live-codex-desktop-voice).