ElevenLabs releases Eleven Realtime speech-to-speech models
The models support over 90 languages and 17,000 voices via an OpenAI-compatible WebSocket API or ElevenAgents.
- Eleven Realtime v1 Max responds in 0.57 seconds, scoring 98.8% on Big Bench Audio reasoning and 87.1% on tool use.
- Eleven Realtime v1 Mini responds in 0.62 seconds, scoring 75.9% on reasoning and 95.8% on Full-Duplex-Bench for conversational dynamics.
- Deployments run through ElevenAgents across phone lines, websites, WhatsApp, and contact center platforms.
Voice agent developers can replace multi-component audio pipelines with speech-to-speech models using existing OpenAI Realtime API integrations.
