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

OpenAI shipped GPT-Live-1 in the API: a single model that listens and speaks at the same time instead of taking turns, so it can parse a user's speech over its own output, catch mid-sentence corrections and respond without the pause and resume rhythm of turn-based voice systems. It delegates reasoning and tool calls to a separate backend model, which OpenAI says cuts interruptions by nearly 80 percent compared with turn-based systems in early tests and adds telephony support for real phone deployments. On the Tau3 benchmark with GPT-6 Astra reasoning behind it, OpenAI reports an 83.6 percent first-attempt task completion rate along with turn-taking, latency and tool-use numbers meant to help engineers compare it against gpt-realtime. For teams building voice agents, this replaces the chained speech-to-text, language-model, text-to-speech pipeline most have been stitching together with a single model built for the handoffs that pipeline used to get wrong.