OpenAI adds three realtime voice and speech models to its API
By Steven Van ·
GPT-Realtime-2, Realtime-Translate and Realtime-Whisper target developers building voice agents, not the ChatGPT app.
OpenAI has released three new realtime audio models in the API behind ChatGPT, aimed at developers building voice agents rather than at the ChatGPT app itself.
- GPT-Realtime-2 is OpenAI's first realtime voice model with GPT-5-class reasoning, built to handle harder requests and carry a conversation forward naturally.
- GPT-Realtime-Translate translates speech from more than 70 input languages into 13 output languages while keeping pace with the speaker.
- GPT-Realtime-Whisper is a streaming speech-to-text model that transcribes speech live as the speaker talks.
GPT-Realtime-2 adds several features for production voice agents: short "preamble" phrases like "let me check that" while it works, parallel tool calls that stay audible ("checking your calendar"), and recovery phrases such as "I'm having trouble with that right now" instead of failing silently. Its context window grows from 32K to 128K tokens for longer, more complex sessions, and developers can now set reasoning effort to minimal, low, medium, high, or xhigh, with low as the default.
Against GPT-Realtime-1.5, OpenAI reports GPT-Realtime-2 scores 15.2% higher on Big Bench Audio at the high reasoning setting, and 13.8% higher on Audio MultiChallenge at xhigh. Full details are in the original announcement.
