OpenAI launches real-time voice model with enhanced features for developers

OpenAI has launched GPT-Live-1, a real-time voice model integrated into its API, enabling developers to build applications with advanced voice interaction capabilities. The model supports full-duplex conversations, allowing simultaneous listening and speaking while handling interruptions, pauses, and background noise. It is designed for phone-based scenarios such as restaurant reservations and customer service, and can be connected to tools like Codex and OpenAI Presence for enterprise integration. GPT-Live-1 reduces latency by combining speech recognition and synthesis within a single model, and allows complex reasoning and tool calls to be handled by backend text models. Tony Stoyanov, co-founder and CTO of a healthcare platform, noted that using GPT-Live-1 reduced code by 80% and cut approximately 23,000 lines. The model outperformed GPT-Realtime-2.1 by 30 percentage points in Full Duplex Bench tests and ranked first in Tau3 end-to-end tasks when paired with GPT-6 Astra. Pricing for the frontend voice layer is $0.05 per minute, with additional costs for backend models and tools. OpenAI also expanded its voice options, adding 12 new voices and plans to add more in the coming months.