OpenAI launched GPT-Live on July 8, 2026, replacing the turn-based voice mode on ChatGPT. The new model listens and speaks at the same time, making conversations feel more natural.

GPT-Live-1 is the premium tier for paid plans. GPT-Live-1-mini powers the free tier. Both run on full-duplex architecture, meaning the AI can interrupt, acknowledge, or stay quiet — just like a human would.

What Full-Duplex Actually Means

Before GPT-Live, ChatGPT’s voice was turn-based. You spoke, then waited for the response. GPT-Live flips that. The AI can acknowledge what you’re saying with “mhmm” or “yeah” while you talk. It can decide to pause, keep listening, or jump in. This happens multiple times per second.

For users, the experience is closer to a phone call than a Q-and-A session.

What’s Missing Today

GPT-Live doesn’t yet support screen sharing or video in ChatGPT. OpenAI is working on those features. API access is also not available yet, though OpenAI opened a notification form for developers and enterprises interested in integration.

The Rollout Timetable

GPT-Live-1 is rolling out now to ChatGPT paid subscribers. GPT-Live-1-mini launches on the free tier this week. No pricing or rate limit details yet for API access.

Voice AI has been awkward for years. GPT-Live makes it feel normal.