Breaking

OpenAI Launches GPT-Live Voice Models That Listen and Speak at Once

July 8, 2026 at 18:30 EDT

Foundation Models
Media Generation

On July 8, 2026, OpenAI unveiled GPT-Live, a new generation of voice models that can listen and speak at the same time, and began a global rollout. It replaces the default ChatGPT Voice experience across iOS, Android and the web. Paid plans (Go/Plus/Pro) receive "GPT-Live-1" as the default, while free plans get "GPT-Live-1 mini." API availability is expected soon (official announcement).

July 8, 2026 · OpenAI

GPT-Live: a voice AI that listens and speaks at the same time

A new full-duplex architecture ends rigid turn-taking — the model decides many times per second whether to speak, listen, interrupt or wait, while frontier models handle hard tasks in the background.

Full-duplex

Processes input & output simultaneously, deciding many times per second

1.5M+

People using voice features every week

iOS · Android · Web

Now the default voice experience in global rollout

Delegation sharply boosts reasoning

Internal evaluations · Advanced Voice Mode vs GPT-Live-1

GPQA — expert-level science reasoning

45.3%

Advanced
Voice Mode

84.2%

GPT-Live-1

BrowseComp — agentic web search

0.7%

Advanced
Voice Mode

75.2%

GPT-Live-1

A lightweight voice model keeps the conversation flowing while complex tasks are delegated in the background to frontier models such as GPT-5.5.

Continuous interaction + delegation

LISTEN & SPEAK

Lightweight voice model backchannels, pauses and interrupts naturally

→

DELEGATE

Hard tasks handed to frontier models like GPT-5.5 in the background

→

RESPOND

Human-like flow returns with no stiff turn-taking or latency

What early users praise

"Smooth, natural, human-like" conversations
Backchannels & natural interruptions
Uninterrupted flow via background delegation
Language learning, hands-free work, long chats

Current limits

Limits on memory and tool access
Gaps with non-native accents & some languages
Video and screen sharing not yet supported
Legacy voice mode still available as an option

Safety-native by design

Voice-native and synthetic-voice testing say GPT-Live matches or exceeds Advanced Voice Mode across key risk areas, with real-time safeguards and teen protections.

Self-harm Emotional dependence Violence Response correction Teen protections

Continue reading

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