OpenAI is preparing another sizable ChatGPT upgrade, and this one looks aimed at the microphone as much as the keyboard. A new bidirectional audio model called Bidi 1 has surfaced inside the app, where it appears designed to let ChatGPT listen and speak at the same time while keeping the flow of a conversation intact.
The move fits OpenAI’s broader push to turn ChatGPT into something closer to a super app, with Codex and AI agents also part of the planned overhaul. That is a smart bit of product stitching: text models have raced ahead, while voice still often feels like a separate, less capable layer bolted on top.
What Bidi 1 adds to ChatGPT voice
Bidi, short for ”bidirectional design,” is meant to handle input and output together rather than forcing the assistant to wait politely for a user to finish. In practice, that means ChatGPT can acknowledge pauses with a quick ”okay”, adapt mid-sentence, and keep track of the full exchange without losing context.
The model is also supposed to be less eager to fill silence with chatter, which is a small detail until you’ve spent time wrestling with voice assistants that treat every pause like a crisis. TestingCatalog says Bidi 1 has already started rolling out to some users, and a public launch could follow this week.
How the new model shows up in the app
According to the code spotted in the app, Bidi 1 sits alongside the standard and advanced model options. When selected, the familiar speech bubble icon turns yellow, which is a neat little visual tell that OpenAI is trying to make the feature feel distinct without turning it into a separate product.
It can also switch tasks on the fly. One example described in the app shows the model counting to ten, then adjusting when interrupted and asked to change the count. That kind of interruption handling is the real test for voice AI, because most systems still behave like brittle phone menus wearing a chatbot costume.
Why OpenAI is pushing ChatGPT voice harder
The timing makes sense. Rivals have been improving multimodal assistants and real-time voice features, and OpenAI has an obvious incentive to make ChatGPT’s voice mode feel as natural as its text interface. The company has not officially announced Bidi 1, and it has also not shared details about GPT 5.6, so for now the rollout is being read from the app itself rather than from a polished keynote.
If OpenAI gets this right, voice could become the default way many people use ChatGPT: faster, more conversational, and a lot less robotic. The bigger question is whether Bidi 1 is a preview of that future or just the first patch in a longer cleanup of voice features that have lagged behind the rest of the product.