OpenAI is preparing a bidirectional voice model called "Bidi 1" (GPT-Bidi-1) for ChatGPT's Voice Mode that can keep speaking while continuing to listen as a user talks, according to internal tests and code/UI traces. A rollout to Codex is also in view.
Early Testing · OpenAI Voice
"Bidi 1" Lets ChatGPT Talk and Listen at the Same Time
A bidirectional (full-duplex) voice model reportedly in early testing can keep speaking while you interrupt — aiming to end the freezing, rigid turn-taking of today's Advanced Voice Mode.
Full-
Processes incoming and outgoing audio simultaneously
3
Intelligence tiers: Instant / Medium / High
23
Counted continuously in one test, self-correcting naturally
Turn-based vs. Bidirectional
Why this matters: today's voice mode can only do one thing at a time.
Current Voice Mode
Freezes when you interrupt
Bidi 1
Adjusts in real time, no freeze
What Bidi 1 Reportedly Adds
Keeps talking while still listening — even as you speak
Switches between tasks back and forth mid-conversation
Better handling of interruptions and pauses
Stronger context retention and memory
Imagined use: "writing code while talking" via Codex
Praised
Largely positive early reactions: adjusting responses without freezing during interruptions, and natural self-correction. Hopes it fixes the rigidity of turn-taking.
Cautious
Rollout is limited; full assessment premature. Worries over weaker non-English performance (Mandarin, African languages) and the tentative "Bidi 1" name lingering in the UI.
All specifications tentative. No official word yet on benchmarks, pricing, or release timing. EEA, UK and Switzerland are expected to receive it later.
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article…