OpenAI GPT-Realtime-2 — Voice Intelligence with GPT-5-Class Reasoning

OpenAI released GPT-Realtime-2 on May 7, 2026, the first voice model built on GPT-5-class reasoning. It leads the Artificial Analysis Big Bench Audio benchmark at 96.6%, expands context to 128K tokens, adds MCP server support and phone calling via SIP, and exits beta to general availability. Two companion models — GPT-Realtime-Translate and GPT-Realtime-Whisper — complete the trio. We review benchmarks, pricing, latency, and what the reasoning upgrade means for production voice agents.