OpenAI brings GPT-Live-1 to its API with full-duplex real-time voice

OpenAI announced it is bringing GPT-Live-1 to its API, letting developers build apps and workflows with real-time voice interaction. The model supports full-duplex conversation, meaning it can listen and speak at the same time, and handle interruptions, pauses, and background noise during a conversation.

GPT-Live-1 combines voice understanding and voice output in one model, cutting down the usual back-and-forth steps to reduce latency. It also works for phone scenarios, such as restaurant bookings and customer service, and can connect with tools like Codex.

Early tests showed that language learning platform Speak saw nearly 80% fewer false interruptions during learners' thinking pauses. A medical service team cut its code by 80%, removing about 23,000 lines. On the Full Duplex Bench, GPT-Live-1 scored 30 percentage points higher than GPT-Realtime-2.1.

GPT-Live-1 is now available in the API, with the front-end voice layer priced at $0.05 per minute. Back-end model and agent tool costs are billed separately. OpenAI also added 12 new voices, including Quartz, Ripple, Vesper, and others, with more voice and language support planned.