OpenAI appears to be testing a major upgrade to voice conversations in ChatGPT. Dubbed by users as “GPT Bidirectional Voice Mode” or “gpt-bidi-1,” the new experience allows the AI to interact in real time, interrupt naturally, count alongside users, and even correct mistakes while a conversation is still in progress.
The feature is currently being spotted by select users in the ChatGPT app and could represent one of the biggest advancements in AI voice interaction since the launch of GPT-4o’s Advanced Voice Mode.
A More Natural Two-Way Conversation
Unlike traditional voice assistants that wait for users to finish speaking before responding, the new bidirectional voice experience enables real-time back-and-forth communication.
A 57-second screen recording circulating online demonstrates the AI counting bananas together with a user. Rather than waiting for the user to finish, the model actively participates in the counting process, responds instantly, and even corrects mistakes during the interaction.
In one example, the AI says:
“Eight… actually, that’s seven.”
This ability to recognize and correct errors while the user is still speaking creates a conversation flow that feels much closer to talking with another person.
GPT Bidirectional Voice Mode is being rolled out in the ChatGPT app!
The model can interrupt the user, count as you speak and it can even correct you when you make a mistake. pic.twitter.com/L8xjKFkQDD
— XIVIX (@XIVIX_134) June 21, 2026
What Makes Bidirectional Voice Different?
The new voice mode appears to build upon OpenAI’s existing GPT-4o Advanced Voice Mode, which already introduced more natural speech patterns, emotional expression, interruptions, laughter, and breathing sounds.
However, bidirectional voice takes things a step further by allowing simultaneous participation during conversations.
Key capabilities demonstrated so far include:
- Real-time interruption when appropriate
- Counting objects alongside the user
- Correcting mistakes instantly
- Dynamic conversation flow
- Faster conversational responses
- More natural turn-taking behavior
Instead of waiting for a clear pause, the AI can react as conversations unfold naturally.
Users Are Calling It “gpt-bidi-1”
The unofficial name “gpt-bidi-1” has emerged from the ChatGPT community, referencing the model’s apparent support for bidirectional audio communication.
While OpenAI has not officially announced a model with this name, users who have gained access to the feature report behavior that differs significantly from existing voice modes.
The naming reflects growing excitement among AI enthusiasts who see the feature as a major leap toward truly conversational AI.
Why Users Are Excited
Many early reactions have been overwhelmingly positive.
Users describe the experience as:
- More engaging than traditional voice assistants
- Faster than standard AI voice interactions
- Closer to real human conversation
- More helpful for learning and counting exercises
- Better at keeping conversations flowing naturally
Some users have even compared the experience favorably to customer service phone systems and call-center agents, arguing that the AI feels more responsive and attentive.
The ability to participate actively rather than passively listening is seen as a significant improvement in conversational design.
Not Everyone Is Convinced
Despite the excitement, some early testers have highlighted several limitations.
Critics point to:
- Occasionally tinny audio quality
- Noticeable response delays
- Awkward filler sounds
- Imperfect interruption timing
- Moments where the conversation feels unnatural
These concerns suggest that while the technology is impressive, there is still room for refinement before a broader rollout.
As with previous voice features, OpenAI is likely to continue iterating based on user feedback.
Evidence Found in the ChatGPT App
Adding to speculation, reports indicate that references to bidirectional audio functionality have been discovered within ChatGPT app code.
While these findings do not confirm an imminent public release, they suggest that OpenAI has been actively experimenting with more advanced voice interaction systems behind the scenes.
The company has yet to officially confirm the existence of a separate bidirectional voice model or explain how the feature works technically.
The Bigger Picture for AI Voice Assistants
The emergence of bidirectional voice interactions highlights a broader trend in AI development.
Technology companies are increasingly moving beyond simple voice commands toward AI systems capable of participating in conversations the way humans do.
Future voice assistants may:
- Interrupt naturally when clarification is needed
- Collaborate on tasks in real time
- Correct mistakes instantly
- Maintain more fluid conversations
- Respond with greater emotional awareness
OpenAI’s latest experiment suggests that the industry is getting closer to that vision.
Final Thoughts
ChatGPT’s emerging Bidirectional Voice Mode may be one of the most important voice AI upgrades yet. By enabling real-time participation, natural interruptions, and instant corrections, OpenAI is pushing conversational AI beyond traditional voice assistant limitations.
Although OpenAI has not officially confirmed the model or its rollout plans, the early demonstrations have already generated significant excitement. If the feature expands to more users, it could redefine what people expect from AI voice interactions and bring ChatGPT one step closer to truly human-like conversation.
Keep yourself updated with all the latest AI news by reading our full coverage here.
Please follow us on our Facebook page and X account for all latest and breaking Windows and Microsoft related news.
