No free plan. The cheapest way in is Pay-as-you-go API usage at From $40/1M audio input tokens (gpt-4o-mini-realtime). Consider free alternatives in the automation & workflows category if budget is tight.
The Realtime API supports WebRTC, which is recommended for browser and mobile clients that need the lowest possible latency, and WebSockets, which are better suited for server-to-server integrations where a backend service mediates between users and the API.
Yes. The API includes server-side voice activity detection (VAD) that detects when a user starts and stops speaking, automatically segments turns, and allows users to interrupt the model mid-response, which the model gracefully handles by truncating its current output.
Yes. The Realtime API supports the same tool and function-calling paradigm as OpenAI's other APIs. You can register tools during session configuration, and the model can decide to call them mid-conversation so the voice agent can fetch data or trigger external actions.
The API is multimodal: a single session can accept and produce text, audio, or both. Developers can configure which modalities are enabled and can mix text inputs (for example, system instructions or silent context updates) with streaming audio within the same conversation.
Usage is billed per token with separate rates for audio and text. For the gpt-4o-realtime model, audio input costs $100 per 1M tokens and audio output costs $200 per 1M tokens, while text input is $5 and text output is $20 per 1M tokens. The more affordable gpt-4o-mini-realtime model charges $40 per 1M audio input tokens and $80 per 1M audio output tokens, with text at $2.50 input and $10 output per 1M tokens. Because speech generates more tokens per second than equivalent text, audio-heavy sessions are priced higher, and developers should monitor session duration and output length to control costs.
See OpenAI Realtime API plans and find the right tier for your needs.
See Pricing Plans →Still not sure? Read our full verdict →
Last verified March 2026