Particle.news
Download on the App Store

OpenAI Reportedly Testing GPT-Bidi-1 to Let ChatGPT Speak and Listen at the Same Time

The model would let ChatGPT handle interruptions by enabling mid-sentence pivots and more natural spoken exchanges.

Overview

  • Reports published June 23–24, 2026 say researchers found GPT-Bidi-1 references in ChatGPT’s app code and UI after code surfaced in mid-June and a TestingCatalog report suggested limited tester access.
  • “Bidi” refers to bidirectional audio, meaning the model is designed to listen and speak simultaneously so it can hear a user interrupt and adjust its reply without pausing.
  • Early in-app clues point to a new yellow voice-mode bubble in the interface and selectable intelligence tiers labeled High, Medium, and Instant to trade response depth for speed.
  • OpenAI has not publicly confirmed the model, released technical documentation, or posted an engineering blog about GPT-Bidi-1 as of June 24, 2026.
  • The effort builds on iterative voice upgrades from 2025 to early 2026 and could speed a shift to voice-first interactions if wider tests validate smoother, real-time conversational control for users.