Realtime 2.1-mini

GPT-Realtime-2.1-mini is a distilled reasoning model in OpenAI's Realtime API lineup for low-latency voice and multimodal interactions. It accepts audio, text, and image inputs over WebRTC, WebSocket, or SIP and outputs text or audio, with support for tool use and configurable reasoning. The July 2026 release adds these capabilities to the mini tier while reducing p95 latency by at least 25% via improved caching.