ElevenLabs has launched Eleven Realtime v1, a voice model that listens to speech and talks back directly, replacing the five or so separate pieces developers used to wire together to build a voice agent. It comes in two sizes, Mini and Max, starts speaking in about 0.6 seconds, works in more than 90 languages and is compatible with OpenAI’s Realtime API, so switching can take just a couple of code changes.
What problem does Eleven Realtime solve?
Building a phone or web voice agent used to mean stacking several systems: one to turn speech into text, one to detect when you stop talking, one to manage turn-taking, a language model to decide what to say, and a text-to-speech voice to say it. Each handoff adds delay and another thing to tune. ElevenLabs put the whole stack behind one model ID. As the company put it, you “shouldn’t have to become a configuration expert to build a voice agent that sounds natural,” per StartupHub.ai.
What is the difference between Mini and Max?
- v1 Max: for complex jobs like customer support and lead qualification. About 0.57 seconds to first audio, 98.8% on the Big Bench Audio reasoning test and 87.1% success on tool-using tasks
- v1 Mini: for natural everyday conversation. About 0.62 seconds to first audio, slightly better at handling pauses and turn-taking, but much weaker at reasoning (75.9%) and tools (60.3%)
ElevenLabs says these scores were evaluated by Artificial Analysis. They are company-reported figures, not our own tests.
Does it keep a text record of conversations?
Yes, and that matters for businesses. Instead of a single sealed audio-to-audio system, Eleven Realtime keeps text at every step inside its pipeline. That lets teams see exactly what the agent heard and what it decided to say, and lets ElevenLabs apply guardrails and hide personal data, which is harder with fully fused voice models.
Why does OpenAI compatibility matter?
Many voice apps are built on OpenAI’s Realtime API, such as those covered in our look at OpenAI’s GPT Live 1 for developers. Because ElevenLabs matches its message format, developers can change the endpoint and model name and keep the rest of their code. That makes it easy to test both and pick on price, voice quality or speed.
How much does it cost?
ElevenLabs charges per minute of connected audio plus language model tokens; exact rates are on its pricing page. It works inside ElevenAgents, with phone, web and WhatsApp channels, or directly through the Realtime API over WebSocket. Apps using WebRTC have to switch to WebSocket first.
Who should use it?
Teams building customer support lines, booking agents or sales assistants that need to sound natural and use tools. Any of ElevenLabs’ 17,000+ library voices can be used. Just remember the voice-cloning ground rules in our voice cloning explainer.




