ChatGPT leak reveals GPT Bidi 1 voice model that listens and speaks at the same time

OpenAI Tests GPT Bidi 1 for Enhanced Voice Interactions

OpenAI is quietly rolling out GPT Bidi 1, a new ChatGPT voice model that represents a significant upgrade to the platform's conversational abilities1

1

. The unannounced bidirectional audio model has already begun appearing for a select group of ChatGPT app users, with TestingCatalog spotting references to the feature last week and reporting that a wider release could happen this week1

1

3

3

. Internal code describes it as a "major leap in intelligence" and "the next generation of Voice," signaling OpenAI's commitment to making speech the primary way people access AI rather than text1

1

.

Simultaneous Listening and Speaking Changes the Game

The name Bidi is shorthand for bidirectional design, which enables the AI assistant to speak, hear, and listen simultaneously—much like human conversation1

1

3

3

. This bidirectional communication approach marks a departure from traditional voice assistants that require users to wait for responses. The model sits in the ChatGPT voice settings alongside Standard and Advanced options, with the voice bubble turning yellow once GPT Bidi 1 is selected1

1

3

3

. Users can speak naturally while the system responds in real time, eliminating long pauses and creating more fluid speech interface experiences2

2

.

Natural Voice Conversations With Better Context Retention

GPT Bidi 1 delivers small, natural acknowledgments like "okay" when users pause or slow down, without interrupting the flow of conversation1

1

3

3

. This creates human-like conversations that feel more organic than previous iterations. The model also handles interruptions seamlessly—ask it to count to ten, interrupt to reverse the count, and it adjusts immediately1

1

3

3

. Perhaps most critically, the model maintains conversation context throughout long exchanges rather than dropping earlier information—a weakness that has plagued ChatGPT's current voice stack1

1

. The system also avoids jumping in during longer pauses, allowing users to think without being interrupted1

1

.

Closing the Gap Between Text and Voice

The release of GPT Bidi 1 represents OpenAI's effort to close the gap between its capable text models and its older voice layer1

1

. This matters because the company is betting that speech will become the dominant interface for AI interactions. The timing aligns with OpenAI's broader plans to transform ChatGPT into a superapp, with a major overhaul focusing on the Codex coding tool and agentic AI tools that can perform tasks for users1

1

. While OpenAI has not officially announced GPT Bidi 1 or detailed GPT 5.6, the fact that the feature is already appearing automatically for some users suggests a phased rollout is underway1

1

3

3

. Users won't need to subscribe or take additional steps to access the feature once it becomes available3

3

. The focus on real-time voice conversations positions ChatGPT to compete more effectively as an everyday AI assistant that understands human behavior and responds naturally2

2

3

3

.