ElevenLabs has released two speech-to-speech models that combine listening, turn-taking and spoken replies. Mini targets simpler conversations, while Max is designed for customer service and workflows involving tools. Both can be used through ElevenAgents or a direct interface, with connection requirements differing from applications that currently rely on WebRTC.
ElevenLabs, the speech-technology company, has released Eleven Realtime v1 Mini and v1 Max. The speech-to-speech models combine recognition, turn-taking and spoken replies. Mini targets simpler conversations, teaching and interactive characters, while Max is designed for customer support and multistep work involving tools. Both are available through ElevenAgents and a direct Realtime interface.[1], [2]
The combined design addresses decisions such as how long to wait after a pause and how to respond when a caller interrupts. ElevenLabs says both models support more than 90 languages. In ElevenAgents, teams can select a Realtime model while using the platform’s phone, web, WhatsApp and contact-center channels. Workflows, knowledge bases, testing and monitoring are also platform functions.[1]
The direct application programming interface (API) follows the OpenAI Realtime specification. Existing integrations can change the endpoint and model name, but the transport uses WebSocket only. Applications connected through WebRTC must change that connection first. The direct API runs separately from ElevenAgents; ElevenLabs recommends the platform for teams needing its communication channels and knowledge-base features.[1]
