Intel / article

Realtime

Source: elevenlabs.io
Author: ElevenLabs editorial sitemap
Date: Published Oct 20, 2026

From the Daily Brief

ElevenLabs released Eleven Realtime, a pair of speech-to-speech models that each act as a complete voice agent instead of a chain of separate speech recognition, turn detection, language model and text-to-speech components. Both support more than 90 languages and the 17,000 voices in the ElevenLabs library. Mini targets low-cost conversation such as assistants and tutors, and Max targets support and multi-step workflows. In Artificial Analysis testing, Max answered 98.8% of Big Bench Audio questions correctly and began speaking in 0.57 seconds on average, while Mini scored 75.9%. Mini scored slightly higher on Full-Duplex-Bench, which tests pauses and interruptions, but Max completed 87.1% of tool-use tasks against Mini's 60.3%, which is why ElevenLabs recommends Max for agents that call tools. The Realtime API follows OpenAI's Realtime specification, so OpenAI's SDK works against it unchanged. The limits are practical: it connects over WebSocket only, so WebRTC integrations need to move first, and the API runs separately from the ElevenAgents platform that provides phone channels, knowledge bases and guardrails.

Read the 2026-10-21 Brief →

Key takeaways · AI-distilled

Eleven Realtime ships as two speech-to-speech models: v1 Mini for low-cost conversation and v1 Max for multi-step, process-following work like support and lead qualification. Both cover 90+ languages and the full voice library.
In Artificial Analysis testing reported by ElevenLabs, v1 Max answered 98.8% of Big Bench Audio reasoning questions versus 75.9% for Mini, while starting to speak in 0.57s on average versus 0.62s.
The gap widens on : v1 Max completed 87.1% of tool-use tasks and v1 Mini 60.3%, which is why ElevenLabs recommends Max for tool-heavy agents. On turn-taking (Full-Duplex-Bench) Mini scored slightly higher, 95.8% to 94.1%.
The Realtime API follows OpenAI's Realtime spec over WebSocket, so existing OpenAI SDK code can switch by changing the endpoint and model name. WebRTC integrations must move to WebSocket first.
The standalone API runs separately from ElevenAgents, so teams needing phone or WhatsApp channels, a knowledge base, or monitoring are told to build on ElevenAgents from the start.

Terms in this piece · Glossary

LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
tool use — A model's ability to call external functions — run code, search the web, edit files — instead of only generating text.
guardrails — The checks around a model that block bad inputs and outputs — filters, validators, and permission rules the model itself can't override.

Read the source elevenlabs.io