OpenAI Launches GPT-Live: The Voice AI That Listens, Thinks, and Talks at the Same Time

Every voice AI you’ve used so far — Siri, Google Assistant, even ChatGPT’s Advanced Voice Mode — works in turns. You talk, it listens, it thinks, it responds. There’s a gap between each phase, and that gap is where the “talking to a robot” feeling lives. OpenAI’s new GPT-Live is built to close that gap entirely.

GPT-Live launched this week on a full-duplex architecture, which means it handles listening, speaking, and reasoning simultaneously rather than sequentially. In practical terms: it can interrupt you, pick up on mid-sentence corrections, and respond in real time the way a human conversation partner does.

What Full-Duplex Actually Means

Traditional voice AI is half-duplex: like a walkie-talkie, only one side operates at a time. You finish speaking, the system processes, then responds. Even the best implementations have 1–2 second latency gaps that feel unnatural in conversation.

Full-duplex systems run both channels in parallel. GPT-Live listens to your input stream while simultaneously generating its response — meaning it can catch “actually wait, I meant the other one” mid-sentence and correct course without making you start over. The architecture also means it doesn’t have a “thinking pause” between receiving your message and beginning to reply.

OpenAI claims this makes GPT-Live the first conversational AI that passes the “phone call test” — where someone on the other end genuinely can’t tell if they’re talking to a person or a machine during a short task-oriented call.

Real-Time Translation Built In

GPT-Live supports real-time translation during conversation. You speak in English, someone else speaks in Spanish, and GPT-Live renders both streams into each other’s language with sub-second latency. OpenAI hasn’t published a full list of supported languages, but early documentation shows coverage of the same languages supported by GPT-4o’s existing multilingual voice features — roughly 50+ languages.

The translation use case is where the full-duplex architecture matters most. In a turn-based system, translation introduces compounding delays: wait for Speaker A to finish, translate, wait for Speaker B to finish, translate back. GPT-Live’s parallel processing means both sides hear each other in close to real time.

Live Web Search Mid-Conversation

The third major feature is live web search that activates automatically when GPT-Live detects a query that needs current information. You don’t trigger it manually — it decides mid-sentence when to reach out for fresh data and incorporates the results into its spoken response without pausing.

This addresses one of the consistent frustrations with voice AI: you ask something time-sensitive (“what’s the score of the game right now?”), the AI either hallucinates or says it doesn’t have access. GPT-Live pulls the answer live and reads it back as part of its response.

Task Delegation to Other Agents

GPT-Live can also hand off tasks to other AI agents mid-conversation. Ask it to book a restaurant while discussing dinner plans and it can delegate that specific subtask — finding availability, checking the menu, initiating a reservation — to a specialized agent without interrupting the main conversation.

This is where GPT-Live fits into OpenAI’s broader agentic strategy. It’s not just a better voice interface; it’s a conversational layer that can orchestrate multiple backend agents on the fly. The implication for enterprise use cases (customer service, scheduling, sales) is substantial.

Where to Get It and What It Costs

OpenAI hasn’t released specific pricing tiers for GPT-Live as of this writing. It’s expected to be available through ChatGPT Plus and Team subscriptions initially, with API access for developers following. The current ChatGPT Plus plan runs $20/month — though OpenAI has been moving higher-capability features to the $200/month Pro tier, and GPT-Live may follow that pattern.

If you’re evaluating AI tools for your workflow, our roundup of the best AI productivity tools in 2026 covers the full landscape — including where GPT-Live fits relative to competitors like Google’s Gemini Live and Amazon’s Alexa+. And for context on what this means for Apple’s own Siri ambitions, our coverage of Apple Intelligence in iOS 27 is worth reading alongside this.