ElevenLabs launches Realtime speech-to-speech models

ElevenLabs describes Eleven Realtime as a new class of speech-to-speech models, with each model designed as a complete voice agent for a different use case.[1] The company says Realtime brings speech recognition, turn-taking, the underlying language model and voice together in one end-to-end system.[2] Both models support more than 90 languages and can use any of the more than 17,000 voices in ElevenLabs’ library.[1] The company says its goal is to let developers create natural-sounding voice agents without becoming configuration experts.[1]

ElevenLabs positions v1 Mini for natural conversations at low cost and says it should be sufficient for most use cases; it names assistants, teachers and interactive characters as possible applications.[1] The company positions v1 Max for more complex work, such as customer support, lead qualification and multi-step workflows.[1] Max is intended for agents that need to follow a process, reason through steps and use tools across multiple stages.[1][3] ElevenLabs recommends Max for applications with heavy tool calling, and its documentation says Max costs more than Mini.[3][1]

ElevenLabs says Artificial Analysis independently evaluated the models across three public benchmarks.[1] In the reported question-answering result, Max answered 98.8% correctly and Mini answered 75.9%.[1] Max began speaking in an average of 0.57 seconds, compared with 0.62 seconds for Mini.[1] Mini, however, scored 95.8% on Full-Duplex-Bench, ahead of Max at 94.1%; Max scored 998 in the Speech Agent Arena, while Mini scored 988.[1] On tool-use tasks, Max completed 87.1%, against Mini’s 60.3%.[1] These results point to different strengths rather than a single model leading every reported measure.[1]

Eleven Realtime is available through ElevenAgents and the Realtime API, though ElevenLabs says the models are rolling out gradually.[2][3] Developers using OpenAI’s Realtime API can try ElevenLabs’ models by changing the endpoint and model name while keeping the rest of their code, according to ElevenLabs.[2] That route has a transport requirement: the Realtime API connects over WebSocket only, so existing WebRTC integrations need to move to WebSockets.[1] ElevenAgents supports deployment across phone, web, WhatsApp and contact-center channels, alongside workflows, knowledge bases, guardrails, testing and monitoring.[1] Max is not available in ElevenLabs’ India and Singapore data-residency environments.[3]