Eleven Realtime
ElevenLabs launched Eleven Realtime, speech to speech voice agent models in two sizes, v1 Mini and v1 Max, with an API compatible with OpenAI's Realtime API.
Each model is a complete voice agent that covers speech recognition, turn-taking, the LLM and text to speech, tuned end to end on a cascaded architecture. Both support more than 90 languages and any of the 17,000+ voices in the ElevenLabs library. Developers on OpenAI's Realtime API can try it by changing the endpoint and model name, or select it in ElevenAgents.
- Date
- Tuesday 20 October 2026
- Lab
- ElevenLabs
- Kind
- model
- Access
- closed API
Figures
| Measure | Value | Measured by |
|---|---|---|
| Big Bench Audio accuracy, v1 Max | 98.8% speech reasoning, 1,000 questions | ElevenLabs |
| Big Bench Audio accuracy, v1 Mini | 75.9% | ElevenLabs |
| Average time to start speaking, v1 Max | 0.57 s | ElevenLabs |
| Average time to start speaking, v1 Mini | 0.62 s | ElevenLabs |
| Full-Duplex-Bench, v1 Mini / v1 Max | 95.8% / 94.1% pauses, turn-taking, interruptions, backchannels | ElevenLabs |
| Speech Agent Arena score, v1 Max / v1 Mini | 998 / 988 blind live conversations | ElevenLabs |
| Tool-use task completion, v1 Max / v1 Mini | 87.1% / 60.3% | ElevenLabs |
The Realtime API uses WebSocket only and runs separately from ElevenAgents. Pricing is not stated on the page.
Sources
This record was checked and corrected against its sources on 30 October 2026. How we check