New Delhi: OpenAI is reportedly preparing a major upgrade for ChatGPT’s voice capabilities with a new bidirectional audio model called GPT-Bidi-1. The
new model may revolutionise conversations, making them feel more natural and allowing the AI to listen and respond simultaneously rather than waiting for a turn.
OpenAI is continuing to pivot ChatGPT from a chatbot to a larger AI platform. In addition to developments in its Codex coding tools and agentic AI capabilities, the company is reportedly pouring money into its voice capabilities, shifting the paradigm of user-AI interactions from text to voice. At the moment, OpenAI has not officially revealed GPT-Bidi-1, and the specifics are from code discovery and early testing reports.
GPT-Bidi-1 could enable more natural conversations
Recently, internal code referencing GPT-Bidi-1 was found, according to TestingCatalog. According to the publication, “Bidi” is short for bidirectional, which will allow the AI to listen, understand and speak at the same time rather than take turns talking to each other like normal conversations.
This might make interactions with ChatGPT more of a human experience. Rather than remaining silent during pauses for users to consider the answers, the assistant can give brief affirmations like “okay” without interrupting the discussion. The new model may also not be able to interrupt in longer pauses, which some have found to be a problem with the existing voice mode.
Better at handling interruptions and remembering context
The reported model is also able to respond more naturally when the conversation takes a different turn. For example, if a userasks ChatGPT to count from one to ten but interrupts midway with a new instruction, GPT-Bidi-1 could immediately adapt without restarting the interaction.
TestingCatalog also claims the model can retain context more effectively during longer conversations. It is expected that instead of losing track of previous exchanges, it is able to recall previous exchanges and produce a relevant response that does not become irrelevant during the exchange.
A new voice model may get its own ChatGPT option
GPT-Bidi-1 reportedly appears in a new voice mode selector in ChatGPT, alongside the standard and advanced voice modes. Some users who have tried the feature have reportedly been notified that the voice interface has transformed into a yellow bubble after picking the new model.
The details are compiled from code references, interface discoveries and initial user testing information provided by TestingCatalog and other sources. No technical paper or official confirmation of the capabilities of this model has been released yet by OpenAI.
Early rollout may already be underway
GPT-Bidi-1 is already available in a limited way to ChatGPT app users, suggesting that it could be rolling out completely soon, says TestingCatalog.
As it expands, GPT-Bidi-1 has the potential to make ChatGPT’s speech capabilities a lot closer to OpenAI’s more sophisticated text models. The new model may help bring conversations closer to a human voice and make them feel faster and smoother as the company continues to push voice-first AI experiences.
