ElevenLabs launches Eleven Realtime speech-to-speech model

ElevenLabs announced Eleven Realtime, a speech-to-speech model for real-time voice conversation, on November 3. The model is available in two versions: Eleven Realtime v1 Mini for fast, natural conversation and Eleven Realtime v1 Max for complex tasks.

Eleven Realtime integrates speech recognition, turn detection, turn-taking, an LLM and speech synthesis into a single model selected without separate configuration, according to datanet.co.kr. Both versions support more than 90 languages, including Korean and Japanese, and can use ElevenLabs' library of over 17,000 voices. They also offer natural turn-taking, interruption during user speech and tool calling.

Eleven Realtime v1 Mini is suited to AI assistants, educational tutors and interactive characters in games and entertainment. Eleven Realtime v1 Max is designed for customer support, lead qualification and multi-step tasks that call external tools.

In Artificial Analysis' October 2026 evaluation, v1 Max scored 98.8% on the Big Bench Audio voice reasoning benchmark with 0.57 seconds to start responding. On Full-Duplex-Bench, v1 Mini scored 95.8% and v1 Max scored 94.1%.

Eleven Realtime is available through the ElevenAgents voice and chat agent platform and a Realtime API. The Realtime API is compatible with OpenAI's Realtime API and supports WebSocket connections, but does not support WebRTC at launch.