OpenAI Launches GPT-Live-1 in the API to Bring Full-Duplex Real-Time Voice to Developers

OpenAI has released **GPT-Live-1** in the API, bringing the continuous, full-duplex voice architecture powering ChatGPT Voice directly to developers at a cost of $0.05 per minute for the voice layer.
The model processes incoming audio continuously while generating responses, enabling simultaneous listening and speaking, natural handling of interruptions, and a turn-taking latency of 0.798 seconds compared to 1.41 seconds for GPT-Realtime-2.1. It can seamlessly delegate background tasks, complex reasoning, or actions to backend models such as GPT-6 Astra or third-party options.
In benchmark evaluations, GPT-Live-1 paired with Astra achieved an 86.2% success rate on Tau3 voice tasks (versus 45.7% for GPT-Realtime-2.1) and registered an 87.0% tool-calling pass rate. The API supports telephony integrations, extended conversation sessions, background noise management, and custom pacing across multiple languages and accents.