On July 8, 2026, OpenAI announced GPT‑Live, a new generation of voice model built on a full-duplex architecture, and began rolling it out globally for ChatGPT's Voice feature. The company also said API access for developers and enterprises is planned soon.
July 8, 2026 · OpenAI
GPT‑Live Brings Full‑Duplex Voice to ChatGPT
A new generation of voice model built on continuous, full-duplex processing rolls out globally in ChatGPT's Voice feature — handling interruptions, backchannel cues and background thinking. API access for developers is planned soon.
2
models — GPT‑Live‑1 & GPT‑Live‑1 mini
3
platforms — iOS, Android, ChatGPT.com
Full‑duplex
continuous input & output, multiple judgments per second
Three generations of voice AI — latency & naturalness
Cascaded
high latency, info loss
Turn-based
lower latency, stiff exchanges
Full‑duplex
continuous, natural flow
Taller column = more processing steps / awkwardness. GPT‑Live collapses the pipeline into one continuous stream.
How the conversation keeps flowing
Listen & talk
Backchannels "mhmm", handles interruptions
→
Delegate
Complex tasks sent to GPT‑5.5 in background
→
Keep talking
Conversation continues while thinking
Availability by plan · ChatGPT Voice
GPT‑Live‑1
Go · Plus · Pro plans
GPT‑Live‑1 mini
Free plan
API: coming soon — registration-based. Pricing & context length not yet disclosed. No video or screen sharing yet.
Praise
"Shockingly natural" — a step change in both quality and latency. Beat Advanced Voice Mode on likability, turn-taking, flow and naturalness in human evals.
Criticism
Some found it unnatural — "an overly active listener" that tries to finish your sentences. Gaps may remain for non-native accents and fluency.
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…