OpenAI prepares ChatGPT voice upgrade with Bidi 1 model

OpenAI is building a new voice model called GPT-Bidi-1 that could fundamentally change how ChatGPT handles spoken conversations. The key upgrade: the model can listen and speak simultaneously, rather than taking turns like a polite but slightly awkward dinner guest.

Code and UI elements related to GPT-Bidi-1 were discovered inside the ChatGPT app around June 16, 2026. No official announcement from OpenAI had been made as of June 23, 2026, but the internal preparations suggest the company is serious about making this happen.

What bidirectional audio actually means

In practical terms, this means the model can handle interruptions and modify its responses mid-sentence. If you cut ChatGPT off halfway through an answer to redirect the conversation, the model should be able to pivot without that jarring pause-and-restart that current voice modes suffer from.

The “Bidi” in the name stands for bidirectional, which is the core technical distinction. Current voice sessions predominantly use GPT-4o, a model that was never built from the ground up for real-time audio processing in both directions at once. GPT-Bidi-1 appears to be purpose-built for this exact use case.

One notable visual change: the upgrade reportedly comes with a yellow voice mode bubble, replacing the current interface styling.

Intelligence tiers and the bigger picture

GPT-Bidi-1 isn’t just about smoother conversations. The model is expected to introduce intelligence tiers categorized as High, Medium, and Instant. This mirrors the flexibility that already exists in OpenAI’s text-based models, where users can choose between faster, lighter responses and slower, more thoughtful ones.

Current voice mode already uses GPT-4o as its primary model, with GPT-4o mini serving as a potential fallback option. The introduction of GPT-Bidi-1 would represent a notable advancement over both, suggesting OpenAI views dedicated voice architecture as the path forward rather than simply bolting speech capabilities onto text models.

OpenAI’s voice capabilities have been continuously refined over the past year, with improvements made throughout 2025 and into early 2026. GPT-Bidi-1 has reportedly been under development since early 2026, meaning this isn’t a rushed release but rather the culmination of months of engineering work.

That said, this is still described as early-stage preparation. There’s a meaningful distance between code appearing in an app and a polished product shipping to millions of users.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.