TokenMix Research Lab · 2026-04-24

GPT-4o Realtime Audio API 2026: Setup + Cost Math

GPT-4o Realtime is OpenAI's end-to-end voice-to-voice API — user speaks, model hears, model thinks, model speaks back, all in a single WebSocket session with ~300ms end-to-end latency. Three model variants exist: gpt-4o-realtime-preview (flagship), gpt-4o-mini-realtime-preview (cheaper), and gpt-4o-audio-preview (HTTP-based, non-streaming). Pricing is meaningful: $0.06 per minute of audio input, $0.24 per minute of audio output on the standard variant. This guide covers the WebSocket setup (15 lines of code), cost math at 3 production scales, latency benchmarks vs ElevenLabs Conversational, and when to pick OpenAI Realtime vs Gemini 3.1 Flash Live. TokenMix.ai routes all three variants.

Table of Contents

• Confirmed vs Speculation
• The Three Realtime Variants
• WebSocket Setup in 15 Lines
• Cost Math: 3 Production Scales
• Latency vs ElevenLabs Conversational + Gemini Live
• When to Pick Each
• FAQ

Confirmed vs Speculation

The Three Realtime Variants

gpt-4o-realtime-preview (flagship):

• WebSocket only, voice-to-voice
• 300ms p50 end-to-end
• $0.06/min audio in, $0.24/min audio out, $5/MTok input text, $20/MTok output text
• Best for production voice agents, real-time assistants

gpt-4o-mini-realtime-preview:

• Same WebSocket architecture
• ~20% slower, still <500ms p50
• $0.01/min audio in, $0.04/min audio out (6× cheaper)
• Best for high-volume consumer voice apps

gpt-4o-audio-preview:

• HTTP-based, non-streaming (send audio, wait for audio response)
• 2-8 second full response time
• $0.06/min, same as realtime
• Best for async voicemail processing, voice-based batch tasks

WebSocket Setup in 15 Lines

Minimum viable connection (Python):

import asyncio, json, websockets

async def voice_session():
    async with websockets.connect(
        "wss://api.openai.com/v1/realtime?model=gpt-4o-realtime-preview",
        additional_headers={
            "Authorization": "Bearer $OPENAI_KEY",
            "OpenAI-Beta": "realtime=v1"
        }
    ) as ws:
        # Configure session
        await ws.send(json.dumps({
            "type": "session.update",
            "session": {"voice": "alloy", "instructions": "Be concise."}
        }))
        # Send audio chunks (base64-encoded PCM16 24kHz)
        await ws.send(json.dumps({
            "type": "input_audio_buffer.append",
            "audio": base64_audio_chunk
        }))
        # Receive audio + text deltas
        async for msg in ws:
            print(json.loads(msg).get("type"))

asyncio.run(voice_session())

For production, add: audio VAD (voice activity detection), interruption handling, turn-taking logic, function calling for tool use.

Cost Math: 3 Production Scales

Assume average call: 2 minutes user audio, 3 minutes agent audio.

Small — 1,000 calls/month:

• Realtime: (1000 × 2 × $0.06) + (1000 × 3 × $0.24) = $840/mo
• Mini: (1000 × 2 × $0.01) + (1000 × 3 × $0.04) = 40/mo
• Audio-preview (async): same token cost as Realtime but no streaming UX

Mid — 50,000 calls/month (call center):

• Realtime: $42,000/mo
• Mini: $7,000/mo
• Hybrid route (Realtime for premium, Mini for simple): ~ 5,000/mo

Enterprise — 500,000 calls/month:

• Realtime: $420,000/mo
• Mini: $70,000/mo
• Custom pricing typically negotiated above 1M min/month

Comparison to ElevenLabs Conversational (at $0.30/min blended): ~$30,000/mo at 50K calls. GPT-4o-mini-Realtime is ~4× cheaper than ElevenLabs while similar quality for most business voice use.

Latency vs ElevenLabs Conversational + Gemini Live

End-to-end (user stop speaking → agent starts speaking):

All four cross the "feels conversational" threshold (<500ms). Pick based on cost or voice polish preference.

When to Pick Each

FAQ

Is GPT-4o-Realtime a single model or pipeline?

Single unified model — OpenAI does voice-to-voice in one forward pass, not STT→LLM→TTS pipeline. This is the main latency advantage vs chained approaches.

Can I use Realtime with a proxy/aggregator like TokenMix.ai?

Most aggregators forward WebSocket traffic. TokenMix.ai supports GPT-4o-Realtime via transparent WebSocket proxy. Adds ~20-50ms of latency for routing but simplifies multi-provider fallback.

Does Realtime support function calling / tool use?

Yes, native. Send tools array in session config, model invokes tools mid-conversation, you execute, return results, conversation continues. Works identically to GPT-5.4 function calling.

What's the audio format?

PCM16 at 24kHz, mono. Base64-encoded in WebSocket messages. Most voice SDKs (Twilio, Daily, LiveKit) support this or can transcode.

Can I interrupt the assistant mid-speech?

Yes. Send input_audio_buffer.append any time — the model halts speaking, processes new input, continues. This is what makes the UX feel natural.

How does pricing compare to Whisper + GPT + TTS pipeline?

Pipeline cost at 5-min call: Whisper ($0.03) + GPT-5.4 (~$0.10) + TTS ($0.05) = $0.18. GPT-4o-Realtime: $0.30 blended for same call. Realtime costs 67% more but latency is 3× better and voice tone/emotion is preserved. For user-facing agents, worth it.

Any rate limits I should know?

Tier 4: 20 concurrent WebSocket sessions. Enterprise: negotiable. For high-volume consumer voice apps, ElevenLabs or Gemini may have higher concurrent limits at similar pricing.

Sources

• OpenAI Realtime API Docs
• OpenAI Realtime Introduction
• Voice AI API Comparison — TokenMix
• ElevenLabs Scribe v2 — TokenMix
• Gemini Flash TTS — TokenMix
• GPT-4o Transcribe — TokenMix

By TokenMix Research Lab · Updated 2026-04-24