GPT-Realtime 2

GPT-Realtime-2 is OpenAI's advanced speech-to-speech model featuring GPT-5-class reasoning for low-latency, highly intelligent voice interactions.

OpenAI's GPT-Realtime-2 elevates conversational AI by merging speech-to-speech capabilities with advanced reasoning. Designed for developers building complex voice agents, the model handles intricate multi-turn conversations, follows instructions with high precision, and executes parallel tool calling reliably. Key features include a 128K context window, configurable reasoning effort, and "preambles" (dynamic filler speech that eliminates awkward silence while the model processes complex tasks). It serves as a highly responsive, low-latency engine for real-time voice-to-action workflows.

https://developers.openai.com/api/docs/guides/realtime