OpenAI Realtime API API Pricing & Rate Card

Realtime WebRTC / WebSocket Audio Protocols & Token Burn Optimization

OpenAI's Realtime API provides low-latency speech-to-speech conversational streaming powered by GPT-4o Realtime over WebRTC and bidirectional WebSockets. Eliminating the multi-hop latency of discrete Whisper STT → LLM Completion → TTS pipelines, it achieves response latencies under 300 milliseconds. However, audio tokens are billed at a steep premium: $100.00 per million audio input tokens and $200.00 per million audio output tokens (compared to $5.00 / $20.00 for text). A continuous 15-minute voice session can easily burn over $4.50 in API credits if audio stream buffering and silence suppression are not tightly controlled.

WebRTC Session Initialization in TypeScript

Exchange ephemeral tokens and connect to OpenAI's WebRTC signaling gateway:

// Step 1: Request ephemeral session token from your backend
const sessionRes = await fetch("/api/realtime-session", { method: "POST" });
const { client_secret } = await sessionRes.json();

// Step 2: Establish peer connection
const pc = new RTCPeerConnection();
const audioEl = document.createElement("audio");
audioEl.autoplay = true;
pc.ontrack = (e) => (audioEl.srcObject = e.streams[0]);

const stream = await navigator.mediaDevices.getUserMedia({ audio: true });
pc.addTrack(stream.getTracks()[0]);

const offer = await pc.createOffer();
await pc.setLocalDescription(offer);

const sdpResponse = await fetch("https://api.openai.com/v1/realtime?model=gpt-4o-realtime-preview", {
  method: "POST",
  headers: {
    Authorization: `Bearer ${client_secret.value}`,
    "Content-Type": "application/sdp"
  },
  body: offer.sdp
});

const answer = { type: "answer" as const, sdp: await sdpResponse.text() };
await pc.setRemoteDescription(answer);

Client-Side Voice Activity Detection (VAD) Controls

To prevent silent audio from burning hundreds of dollars in audio input tokens, configure server-side VAD with tight silence thresholds: `{"turn_detection": {"type": "server_vad", "threshold": 0.5, "silence_duration_ms": 300}}`. Additionally, mute client microphone audio tracks automatically when the user is not actively speaking.