Breaking

OpenAI's New Voice Model 'gpt-bidi-1' Spotted in DevMode

June 16, 2026 at 12:22 EDT

Foundation Models
Media Generation

OpenAI appears to be preparing a next-generation voice model for ChatGPT called "gpt-bidi-1" (GPT-Bidi-1), which was spotted within the developer community DevMode. The information is at the leak stage, with no official announcement yet.

Leak · OpenAI

"gpt-bidi-1": OpenAI's Next Voice Model Spotted Before Launch

A revamped ChatGPT voice feature built for more natural, two-way conversation is reportedly arriving "very soon" — though it remains an unannounced, unreleased name with no official specs.

BiDi

Bidirectional voice model in development since ~March 2026

Full‑duplex

Listen and speak at once — backchanneling like "yeah", "oh I see"

Official specs: no pricing, benchmarks or platforms confirmed

From turn-based to continuous: how the pipeline changes

Today — Cascaded Advanced Voice Mode

STT → LLM → TTS Waits for you to finish — largely turn-based

Next — BiDi (gpt-bidi-1)

Continuous audio in ⇄ Continuous audio out Responds instantly to interruptions & changes of direction

The hope

Natural backchanneling for a human feel
Smoother interruptions, near full-duplex flow
An edge for customer support & voice agents

The caveats

Prototype glitched after a few minutes
Original target timeline slipped
Good at talking, still weak at following instructions

Still a leak — not a launch.

"gpt-bidi-1" is an internal code name that may change, with no real-world evaluation yet. With documentation and benchmarks absent, OpenAI's formal announcement remains the milestone for confirming the full picture.

Continue reading

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