OPENAI LAUNCHES GPT-LIVE-1 IN THE API GPT-Live-1 brings ChatGPT’s real-time voice model to developers, allowing AI agents to listen and speak at the same time, handle interruptions naturally, and delegate deeper reasoning or actions to models like GPT-6 Astra. Key details: Full-duplex voice with 0.798s turn-taking latency vs. 1.41s for GPT-Realtime-2.1 86.2% on Tau3 voice tasks when paired with Astra, vs. 45.7% for GPT-Realtime-2.1 87% tool-calling pass rate on spoken requests Supports phone calls, background noise, long sessions and custom tone/pacing Can pair with OpenAI or third-party backend models New voices across more accents, dialects and languages Available today at $0.05/min for the voice layer OpenAI says Speak saw nearly 80% fewer interruptions in early testing.