OpenAI appears to be building a new audio model, tentatively named GPT-Bidi-1, which has surfaced through code references, UI sightings, and limited early user tests on web and mobile clients. Unlike the current Advanced Voice Mode, GPT-Bidi-1 is reportedly built on a bidirectional, full-duplex audio architecture, meaning ChatGPT would speak and listen at the same time rather than waiting for the user to finish before responding. The model is designed to absorb interruptions mid-response, issue short acknowledgements without cutting the user off, and hold longer conversational context than the current voice stack allows. UI elements linked to GPT-Bidi-1 suggest selectable tiers labeled High, Medium, and Instant. Some users in app previews have spotted GPT-Bidi-1 in a model-selector UI alongside the existing Advanced Voice Mode, with the voice bubble reportedly changing colour when the new mode is active. OpenAI has made no official announcement. The move is widely seen as OpenAI's attempt to close the gap between ChatGPT's advanced text capabilities and its comparatively limited audio experience.