Key points
- ElevenLabs launched Eleven Realtime, speech to speech models that act as a full voice agent, in two versions: v1 Mini and v1 Max.
- Both support 90+ languages and any of the 17,000+ voices in the ElevenLabs library; Max targets support, lead qualification and multi-step workflows.
- Artificial Analysis benchmarks: Max answered 98.8% of Big Bench Audio questions correctly, Mini 75.9%; Max completed 87.1% of tool-use tasks, Mini 60.3%.
- The API follows OpenAI's Realtime spec over WebSocket, so existing integrations switch by changing the endpoint and model name.
Why it matters for marketers: Marketers and agencies building voice agents for support or lead qualification can test one model instead of tuning a multi-part stack, and may switch from OpenAI's Realtime API with minimal code changes.
ElevenLabs has introduced Eleven Realtime, a class of speech to speech models where each model works as a complete voice agent. The company says this removes the need to separately choose and tune speech recognition, voice activity detection, turn-taking, the language model and text to speech.
There are two versions. v1 Mini is aimed at natural conversation at low cost, such as assistants, teachers and interactive characters. v1 Max is aimed at harder work like customer support, lead qualification and multi-step workflows. Both cover more than 90 languages and can use any voice in the company's library of over 17,000.
ElevenLabs cites independent testing by Artificial Analysis. On Big Bench Audio, a set of 1,000 spoken reasoning questions, Max scored 98.8% and Mini 75.9%. Max begins speaking in 0.57 seconds on average, Mini in 0.62. On tool-use tasks Max completed 87.1% versus 60.3% for Mini, which is why ElevenLabs recommends Max for heavy tool calling.
For marketing and content teams, the practical shift is lower setup effort for voice agents that handle calls, lead screening or customer questions. Timing details such as how long the agent waits after a pause, and how it handles interruptions, are adjustable settings in ElevenAgents.
Teams already using OpenAI's Realtime API can point their code at ElevenLabs by changing the endpoint and model name. Note that the API works over WebSocket only, so WebRTC setups must move first, and it runs separately from ElevenAgents. Anyone needing phone, WhatsApp, a knowledge base or monitoring should build on ElevenAgents from the start.
Sources
This brief is our own summary. Read the full reporting at the source.
