This post is from a low-reputation account and contains an unverified outbound link. Be cautious before clicking external links.

ElevenLabs Eleven Realtime: a 2-line switch from OpenAI Realtime, and what a call really costs

tokenlean(25)
Published in
#ai
Words
296
Reading
2 min
7h
Listen to this post
~2 min

Disclosure: this post has affiliate links. If you sign up through them I may earn a commission, at no extra cost to you.

ElevenLabs launched Eleven Realtime on Oct 20: two speech-to-speech models behind an API that speaks OpenAI's Realtime protocol. I went through the docs and pricing page so you don't have to. Here's the short version (full guide with the server code: TokenLean).

The two-line switch

If your app already uses @openai/agents-realtime over WebSocket, you change the URL and the model:

const transport = new OpenAIRealtimeWebSocket({
  url: "wss://api.elevenlabs.io/v1/convai/realtime",   // was OpenAI's endpoint
});
const session = new RealtimeSession(agent, {
  transport,
  model: "eleven_realtime_v1_mini",                    // or eleven_realtime_v1_max
});

Your backend mints a short-lived client secret so the API key never reaches the browser.

Mini or Max?

v1 Miniv1 Max
Time to first audio0.62 s0.57 s
Spoken reasoning (Big Bench Audio)75.9%98.8%
Turn-taking (Full-Duplex-Bench)95.8%94.1%
Tool tasks60.3%87.1%
Tokens in / out per 1M$0.20 / $6.00$1.80 / $9.00
Connected audio$0.08/min$0.08/min

Rule of thumb: chatty assistant → Mini. Anything that calls tools → Max.

What a 5-minute call costs

Audio is a flat 5 × $0.08 = $0.40. With my assumptions (30 turns, ~3k tokens of context per turn, short replies), Mini lands near $0.43 and Max near $0.58. On Mini the minutes are the bill; on Max, trimming the system prompt is the lever.

The catches

  1. WebSocket only, so WebRTC apps must switch transport first.
  2. Over WebSocket you capture and play 24 kHz PCM16 audio yourself.
  3. Server-side SDK docs aren't published yet; the guide covers the browser flow.
  4. Above your plan's concurrency limit, minutes bill at $0.16.

The free tier includes roughly 14 minutes of Realtime v1 Mini, enough to point your existing client at it and listen. Try ElevenLabs →


Affiliate disclosure: TokenLean participates in affiliate programs, including ElevenLabs'. Some links here are affiliate links; I may earn a commission if you buy, at no extra cost to you. Facts come from ElevenLabs' public docs and pricing page (checked Nov 3, 2026); check the official pages before you commit.

ElevenLabs Eleven Realtime: a 2-line switch from OpenAI Realtime, a... | Ecency