New Realtime API for speech-to-speech applications

We're launching the Realtime API (beta) — enabling low-latency, speech-to-speech experiences with GPT-4o.

Features

• Natural speech input and output (no separate STT/TTS step)
• Support for function calling during voice conversations
• 6 preset voices
• WebSocket-based streaming

This is currently in beta. Pricing: $5.00/1M input tokens, $20.00/1M output tokens for audio.

Check the Realtime API guide in our docs for WebSocket connection examples and best practices.