OpenAI has new voice models that reason, translate, and transcribe as you speak

OpenAI launched three new realtime voice models for reasoning, live translation, and streaming transcription.

GPT-Realtime-2 handles live conversation reasoning, while GPT-Realtime-Translate converts speech from 70+ languages into 13 languages.

GPT-Realtime-Whisper provides low-latency streaming transcription as speech happens.

The models are available through the Realtime API with usage-based pricing and can be tested in Playground.

The release gives developers new tools to build faster voice apps with reasoning, translation, and speech-to-text.