GPT-Live-1 Is Now Available to Developers Through the OpenAI API

GPT-Live-1 has launched in the OpenAI API, giving developers access to the same voice technology that powers the latest ChatGPT Voice experience. The model focuses on natural real-time conversations where users can interrupt, pause, and speak while the AI continues processing audio.

Unlike traditional voice assistants that alternate between listening and speaking, GPT-Live-1 supports full-duplex conversations, allowing it to listen and respond at the same time.

GPT-Live-1 costs $0.05 per minute

The front-end voice layer costs $0.05 per minute. Developers must separately pay for backend models, tools, or other services that GPT-Live-1 uses during a conversation.

GPT-Live-1 handles the real-time interaction layer, including speech, listening, pauses, interruptions, and deciding when it needs additional help.

Developers can delegate more demanding reasoning or tool calls to another backend model while GPT-Live-1 keeps the conversation moving. It can also work with third-party models, agent frameworks, and developers’ own services.

This architecture means a voice agent can continue talking with a user while another model or service completes a task in the background, reducing lengthy pauses during complex requests.

OpenAI separates conversation from heavier reasoning

GPT-Live-1 differs from existing Realtime audio models by separating the conversational voice layer from more demanding reasoning and tool execution.

OpenAI says the model improves interruption handling by processing incoming and outgoing audio together. Developers also get greater control over tone, speaking speed, and conversational style through system prompts.

The model can better handle silence and background noise without unnecessarily interrupting users, while improved context retention should help during longer conversations.

GPT-Live-1 also supports telephony integrations, opening the door to phone-based AI agents.

OpenAI claims major benchmark gains

According to OpenAI, GPT-Live-1 improves the Full Duplex Bench score by 30 percentage points compared with GPT-Realtime-2.1.

When paired with GPT-6 Astra at medium reasoning effort, OpenAI says the system ranks first on Tau3, an end-to-end benchmark designed to evaluate voice agents.

OpenAI has also introduced 12 new real-time voices: Quartz, Ripple, Vesper, Willow, Stone, Gleam, Meridian, Bossa, Tempo, Beacon, Delta, and Cinder. They expand support across additional accents, dialects, and languages.

The launch arrives during a busy period for OpenAI. The company recently paused its $200 Pro plan following overwhelming demand, while GPT Images 2.5 has also been released.

GPT-Live-1 also brings API developers closer to the voice experience already available through ChatGPT Voice, but with the flexibility to connect voice interactions to their own models, agents, and services.