OpenAI brings GPT-5-level reasoning to its speech models

OpenAI launched three new speech models: GPT-Realtime-2, GPT-Realtime-Translate, and GPT-Realtime-Whisper.

GPT-Realtime-2 adds stronger reasoning and a larger context window for more capable voice agents.

GPT-Realtime-Translate is built for real-time multilingual translation, while GPT-Realtime-Whisper focuses on faster streaming transcription.

The release broadens OpenAI’s speech API with more specialized tools for natural, real-time audio apps.