Breaking

Tests Point to Real-Time Translation in OpenAI's New 'Bidi 1' Voice Model

June 23, 2026 at 05:26 EDT

Foundation Models
Media Generation

Test information is spreading that OpenAI's reportedly in-development next-generation bidirectional voice model, "Bidi 1" (gpt-bidi-1), will support real-time translation. OpenAI has not confirmed any official release or naming, and what is currently reported is limited to UI traces and prototype videos on ChatGPT Voice Mode.

Unofficial · Test-Stage Leak · OpenAI

OpenAI's 'Bidi 1' Wants Voice That Listens While It Talks

A reportedly upcoming voice model — GPT-Bidi-1 — is built on a bidirectional audio architecture for ChatGPT Voice mode, aiming to handle interruptions, pauses, and real-time translation more like a human conversation.

70+

Input languages on the existing GPT-Realtime-Translate path

Output languages for live speech translation

2-way

Bidirectional audio: speak and listen at the same time

Turn-based vs. bidirectional conversation

You speak

↓ wait ↓

It replies

Current Voice mode
One at a time — weak on interruptions

You speak ⇄ it listens

⇅ overlap ⇅

It speaks ⇄ it hears you

Bidi 1 (reported)
Interrupt, pause, resume mid-sentence

What's reported about Bidi 1

Name

GPT-Bidi-1 — final name may change

Behavior

Listens while speaking, switches tasks mid-sentence, flexible pause/resume

Availability

ChatGPT Voice mode first, eventually the API

Caveat

Still a speaking-length limit; smooth only up to a certain range

Positive reception

"Conversation feels closer to human"
Flexible interruption and pause handling
Addresses Voice mode's core weaknesses

Cautious view

Still in testing — no official announcement
No confirmed benchmarks, pricing, or release date
Wait for an official release before judging

The takeaway — Positioned as a next-generation model after GPT-Realtime-Translate, Bidi 1 pushes toward more natural, two-way voice. Once anything official lands, competition in real-time voice interaction is expected to accelerate.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.