Full-Duplex Audio Lets AI Listen While Speaking, Breaking the Turn-Taking Barrier

Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, announced interaction models on 11 May 2026 using full-duplex audio processing—simultaneous speaking and listening—rather than the sequential turn-taking of current voice assistants. The TML-Interaction-Small model responds in 400 milliseconds, triple the speed of OpenAI's GPT-realtime-2.0. Full-duplex architecture enables natural interruptions and clarifications, shifting voice AI from tool-like interactions toward conversational rhythm.