OpenAI is reportedly preparing a major upgrade to ChatGPT's voice mode: a bidirectional, full-duplex audio model that can listen and speak at the same time. Internal references tag it "Bidi" (some code strings read GPT-Bidi-1), and the feature has begun surfacing in the ChatGPT interface for a subset of users. OpenAI has not officially announced it as of June 23, 2026, so treat this as a reported rollout still in limited testing, not a shipped product.
What This Enables for Creators
If the reports hold, a full-duplex voice model changes how you talk to ChatGPT during creative work. You could brainstorm a script, shot list, or song structure out loud while your hands stay on the keyboard or camera, interrupting and redirecting the model mid-thought instead of waiting for it to finish. Early testers describe reported real-time translation, which would help creators caption or localize spoken content on the fly. Nothing here is live for everyone yet, so the concrete action today is to watch your ChatGPT voice selector for a new "Bidi (Latest)" option and test it the moment it appears in your account.
Why It Matters
Today's Advanced Voice Mode still works in strict turns: it freezes the instant you say "mm-hm" and cannot speak and listen at once. A bidirectional model closes that gap toward natural conversation, where the assistant can acknowledge, interrupt, and correct in real time. As early hands-on reports note, that responsiveness is the difference between dictating to a tool and actually talking with one, which matters for hands-free creative workflows and accessibility.
Key Details
According to the code references spotted in the ChatGPT app, Bidi can speak while listening, offer brief acknowledgments without cutting the user off, switch tasks mid-conversation, retain context across an entire session, and reduce false starts during long pauses. It reportedly appears in the model selector with a yellow voice bubble and offers High, Medium, and Instant intelligence tiers that mirror ChatGPT's text side. Users would likely keep toggling between the new Bidi option and the current Advanced Voice Mode rather than being switched over wholesale. As multiple outlets reported on June 21 to 23, there are still no official benchmarks, documentation, or confirmation from OpenAI.
What to Do Next
Open the ChatGPT mobile or web app and check whether a "Bidi (Latest)" entry has reached your voice selector yet. If it has, run a short hands-free test: start a voice session, interrupt the model mid-sentence, and see whether it adapts without resetting. Keep your existing Advanced Voice Mode workflow as the fallback until OpenAI confirms the feature and publishes tier availability, since a reported rollout can change names or get pulled before a wide launch.
