OpenAI is developing a bidirectional audio model that continuously processes sound in the background and can instantly recognize user interjections, adapting its responses on the fly. This technology enables natural dialogue, allowing the system to handle interruptions or changes in thought mid-sentence. Such flexibility is crucial for complex scenarios like virtual support agents that must adapt to evolving conversation contexts without losing track.

The prototype is currently under testing, with some issues observed during long sessions, leading to a delay of the public release until at least the second quarter of 2026.

Source: theinformation.com