OpenAI has launched GPT-Live-1 in the API, delivering natural, full-duplex voice conversations to developers. This model enables simultaneous listening and speaking, creating more natural voice experiences for apps and business workflows. A core strength is smooth interruption handling, which early evaluations show can cut interruptions by almost 80% compared to previous turn-based systems.

The API focuses on new capabilities allowing developers to steer voice experiences around specific users and goals. Key features include improved interruption handling via a single reasoning model, reasoning and tool calling delegation to backend models, and the ability to shape an agent's tone, pace, and style through system prompts. The system also offers better silent context management and background noise handling without interrupting the flow.

Telephony support enables deployment of full-duplex voice agents for phone calls, from reservations to customer support. GPT-Live-1 natively provides ASR transcripts and response text, supporting strong alphanumeric understanding and keyword biasing. While not a turn-based model, it supports turn detection.

This simplifies voice-agent architecture by reducing latency and coordinating fewer stages, allowing developers to pair the model with various backend tools for scheduling, order updates, or complex customer issues. The launch represents a significant step toward more natural, efficient voice interactions in API-driven applications.