OpenAI is laying the groundwork for a new voice model that could bring a major update to ChatGPT’s spoken conversations. The model, tentatively identified as GPT-Bidi-1, has been seen in preparations across both web and mobile, suggesting the company may be close to introducing it to users.
The name appears to reference a bidirectional, or BiDi, architecture that OpenAI has reportedly been developing since earlier this year. In practical terms, that kind of system is meant to handle conversation more fluidly. Rather than waiting for a user to finish speaking, it can listen and respond at the same time, adapt to interruptions, and adjust its output without stalling mid-response.
The update matters because OpenAI has made faster progress in text than in voice. Its text models have advanced to the GPT-5.5 generation, while ChatGPT’s voice capabilities have remained tied to an older audio system. That has created a noticeable divide between the assistant’s performance in writing and in spoken dialogue. A stronger voice model would help narrow that gap and better support OpenAI’s wider push to make speech a primary interface for AI.
That strategy is already visible in the company’s broader plans, including audio-focused hardware ambitions and voice-based support tools. GPT-Bidi-1 is positioned as part of that effort, with the promise of more natural interactions and a meaningful improvement in reasoning.
Based on the current signs, ChatGPT users may not be forced into the new experience immediately. Instead, the app is expected to keep the existing Advanced Voice Mode while adding a new Bidi, or Latest, option alongside it. That would allow users to choose between the two modes rather than replacing one with the other outright.
OpenAI also appears to be considering multiple intelligence settings for the new voice experience. Those options, described as High, Medium, and Instant, would mirror the tiered choices already available in text-based use. Such a setup would let users balance speed against depth depending on the task, whether they want a quick reply or a more thoughtful answer.
A recent interface change may already be connected to the redesign. ChatGPT’s voice bubble can now be dragged to the center of the screen, which could be an early signal that the interface is being prepared for a refreshed voice experience.
The company has not publicly confirmed a launch date, and it remains unclear whether the rollout will begin immediately or later. Still, the appearance of GPT-Bidi-1 across multiple surfaces suggests the project is moving beyond internal testing.
For OpenAI, the upgrade would be about more than improving audio quality. It would be a step toward bringing voice in line with the company’s latest model capabilities, while making ChatGPT feel more responsive in real-time conversation. If the rollout happens soon, it could mark one of the biggest changes to ChatGPT’s voice mode in months.