OpenAI launches GPT-Realtime-2, GPT-Realtime-Translate, and GPT-Realtime-Whisper audio models for developers building voice applications.

OpenAI introduced three audio models in the API: GPT-Realtime-2 (a voice model with GPT-5-class reasoning), GPT-Realtime-Translate (supporting 70+ input languages and 13 output languages for live translation), and GPT-Realtime-Whisper (a streaming speech-to-text for live transcription).

GPT-Realtime-2 increases context window from 32K to 128K to support longer sessions, adds adjustable reasoning effort (minimal, low, medium, high, xhigh), and includes preambles and parallel tool calls that make agent actions audible to users during task completion.

GPT-Realtime-2 (high) scores 15.2% higher on Big Bench Audio for audio intelligence than GPT-Realtime-1.5; GPT-Realtime-2 (xhigh) scores 13.8% higher on Audio MultiChallenge for instruction following. Zillow reported a 26-point lift in call success rate after prompt optimization (95% vs. 69%) on adversarial benchmarks.