gpt-realtime is OpenAI's new speech-to-speech model for production voice agents, delivering low latency and natural, expressive speech. The Realtime API is now GA, adding key features for developers like remote MCP support, image input, and SIP phone calling.