GPT Realtime 2

GPT-Realtime-2 is a next-generation audio model released by OpenAI on May 7, 2026, as part of the Realtime API. It combines GPT-5-class reasoning, parallel tool calls, and a context window expanded to 128K tokens (up from 32K in the previous version). A new "preamble" feature lets the model speak short acknowledgement phrases ("let me check that", "one moment") before generating a full response, along with audible announcements of tool calls in progress.

On OpenAI benchmarks, GPT-Realtime-2 (high) scores 15.2% higher than its predecessor GPT-Realtime-1.5 on Big Bench Audio (audio reasoning) and 13.8% higher on Audio MultiChallenge (multi-turn conversation). Early tester Zillow reported a 26-point increase in call success rate (95% vs. 69%) after prompt optimization. The model is accessible via WebRTC, WebSocket, and SIP, with full EU Data Residency support.