OpenAI Realtime API

OpenAI Realtime API gives developers a general-purpose realtime model that accepts and produces audio and text, follows instructions, and calls tools during a conversation. It supports WebRTC, WebSocket, and SIP connections, making it a foundation for browser, app, and phone voice agents rather than a packaged receptionist.

Editorial research · Reviewed Jul 10, 2026

Test the agent live

Official Realtime Playground; an OpenAI Platform account and eligible API access may be required.

Open the live demo

Quick facts

Platform type

Developer API

Inbound calls

Yes

Outbound calls

Yes

Human handoff

Yes

Setup difficulty

Technical

Pricing model

$32/M audio input tokens and $64/M audio output tokens for gpt-realtime

Developer friendly

Yes

Pricing details

OpenAI lists gpt-realtime at $32 per million audio input tokens and $64 per million audio output tokens, with separate text-token rates. Telephony, storage, observability, and other services are additional. Verify the current model page before estimating per-minute cost.

Integrations

WebRTCWebSocketSIPOpenAI Agents SDKCustom toolsCustom API

Data and permissions

Voice agents may talk to customers, access records, record calls, and trigger external actions.

Can read customer data

Can write or update external systems

Can send messages

Can trigger workflows

Requires API keys

Records calls

Human approval available

Use scoped API keys and server-side tool authorization. Disclose AI interaction, prohibit deceptive impersonation, obtain consent for recording or custom voices, and do not expose sensitive tools directly to model-generated arguments.