OpenAI tests "Bidi 1," a bidirectional ChatGPT voice model that can speak and listen simultaneously
Story Overview
OpenAI is running early internal tests of an unreleased bidirectional voice model nicknamed Bidi 1 inside ChatGPT that supports simultaneous speaking and listening, allowing the system to handle interruptions, switch tasks mid-sentence, and keep context without freezing the conversation.
Early tests already show uninterrupted counting
Short demonstrations reveal the model counting forward and backward in one fluid pass, with no artificial pauses while the user speaks over it.
Release timing and access stay unknown
No official word exists yet on when Bidi 1 might reach ChatGPT users, which plans could get it first, or whether it will appear in other OpenAI products.
Positive users praise OpenAI's bidirectional voice model Bidi 1 for ChatGPT as sounding impressively human and long-awaited, while negative users dislike its name, interruptions, or occasional robotic delivery.
No Digg Deeper questions have been answered for this story yet.
Most Activity
All voice models are bi - either bimodal or bidirectional.
OpenAI’s new upcoming „bidi“-voice mode sounds insane!
@kimmonismus Voice models going full real-time dialogue feels like the biggest UX shift since chat itself
@kimmonismus i know its bidirectional, but horrible name “ bidi “ 😭 it gives me completely different vibe
@kimmonismus I’m not fond of being interrupted. Not sure this is something I want in my voice mode experience.
@kimmonismus i think i had that already a few times randomly as a/b test. it sounded kinda "drunk" a few times and kinda felt worse...
@kimmonismus Isn't this unsafe?
Like what if an autist mistakes this for a real girlfriend who tells them to invest all their money in Bitcoin, vote Democrat and jump in front of a moving train?
@kimmonismus Am I going to have to stop using WisprFlow now?
@nisxant Hahahaha valid
@kimmonismus anything but 5.6 atp
@kimmonismus Can ChatGPT Bidi delegate a task to a thinking/reasoning model in the background while continuing the current voice/chat conversation?
@kimmonismus Is it GPT 5.6??
@kimmonismus Built live audio/video before this wave. Full-duplex is where voice gets real and brutally hard. The model is the easy part. The work nobody demos is handling interruptions and sub-200ms turn-taking so the agent never talks over you. Glad to see OpenAI go after it.
@kimmonismus Low latency duplex audio is going to change how we interact with software forever.
@kimmonismus wispr flows days are counted
@kimmonismus Everyone will mock the name until they use it once and realize normal voice assistants feel like voicemail from 2009.
@kimmonismus AI that interrupts😅 Don't we have too many people in life that do the same thing?
@Rawlala1 @kimmonismus mama, this model is said to be released together with 5.6 and 5.6 pro so
@kimmonismus Wow that's awesome I have always want this feature
@kimmonismus This is what they're going to use in their device I guess
@kimmonismus Can we hope for today or...?