Build more natural voice experiences with GPT‑Live‑1 in the API

OpenAI has announced the API launch of GPT-Live-1, a voice model designed to enable natural, full-duplex conversations for applications and business workflows. Previously introduced in ChatGPT, the model is capable of listening and speaking simultaneously. According to OpenAI, GPT-Live-1 reasons over incoming and outgoing audio together in a single model, avoiding the latency and handoff challenges associated with traditional chained speech-to-text, reasoning, and text-to-speech architectures. Early evaluations indicate that the model's smooth interruption handling reduced interruptions by nearly 80% compared to previous turn-based systems. GPT-Live-1 allows developers to delegate deeper reasoning and tool calls to backend text models, such as GPT-6 Astra or third-party models. The API release introduces features including tone, pace, and style steering through system prompts, silent context management, background noise handling, and long-session reliability. Additionally, the model features native telephony support for phone call deployments and provides ASR transcripts, response text, and keyword biasing alongside native turn detection. OpenAI reports that GPT-Live-1 improves Full Duplex Bench performance by 30 percentage points over GPT-Realtime-2.1. The front-end voice layer is available in the API at $0.05 per minute, with a broader selection of voices across accents, dialects, and languages. Enterprises can also utilize OpenAI Presence to deploy trusted voice agents powered by the model.