Low-latency speech-to-speech API for building natural voice agents with GPT-4o.
The Realtime API streams audio both ways with GPT-4o, so users get natural, interruptible conversation instead of press-and-wait prompts.
Who it's for: Developers building phone agents, in-app voice assistants, and accessibility tools that need human-like latency.
Stream audio in and out with near-human turn-taking.
Trigger APIs mid-conversation without breaking the flow.
Send images and get spoken responses.
Token and audio streaming keep responses snappy.
The OpenAI Realtime API is the most natural voice-agent backend in 2026 — best-in-class latency, as long as you can stomach the per-minute cost.