GPT-Realtime-2 - MixRoute API

GPT-Realtime-2 is OpenAI’s reasoning-capable realtime voice model for speech-to-speech agents and tool use. Use the Realtime WebSocket endpoint. The Chat Completions endpoint is not supported for this MixRoute route.

Key capabilities

Realtime speech - Text and audio input and output
Configurable reasoning - Adjust reasoning effort for voice workflows
Tool use - Supports function calls in realtime sessions
Multimodal input - Accepts text, audio, and image input

Quick example

Node.js WebSocket

Install the ws package, then connect with the model query parameter and a Bearer token.

import WebSocket from "ws";

const ws = new WebSocket(
  "wss://api.mixroute.ai/v1/realtime?model=gpt-realtime-2",
  { headers: { Authorization: "Bearer YOUR_API_KEY" } },
);

ws.on("open", () => {
  ws.send(JSON.stringify({
    type: "response.create",
    response: { modalities: ["text"] },
  }));
});

ws.on("message", (data) => console.log(data.toString()));

Parameters

Parameter	Type	Required	Description
`model`	string	Yes	Must be `gpt-realtime-2`.
`Authorization`	header	Yes	Bearer API key sent during the WebSocket handshake.
`session.update`	event	No	Configure modalities, instructions, tools, audio, and reasoning.
`response.create`	event	Yes	Start model generation for the current session.

Vendor reference

OpenAI GPT-Realtime-2 model page