GPT-Live-1 collapses listening and speaking into one model rather than chaining separate STT, reasoning, and TTS stages. It powered ChatGPT Voice from July 8, 2026, and OpenAI Developers announced API access on September 10, 2026 so builders can pair it with their own backend model and tool-calling harness. HeyGen's same-day open-source LiveAvatar demos use it as the brain-and-voice layer under a real-time avatar face.