ElevenLabs Launches Realtime Voice Agents That Sound Human

ElevenLabs, a company known for its realistic AI voices, has launched Eleven Realtime, a new product that makes it much easier to build voice agents—AI that can talk with you naturally. Normally, developers have to piece together several different technologies to make a voice agent work: speech recognition, voice activity detection, turn-taking, a language model, and text-to-speech. That's a lot of work. Eleven Realtime bundles all of that into one model, so developers can focus on the conversation experience instead of the plumbing.

The new models come in two versions: v1 Max and v1 Mini. Both support more than 90 languages and can use any of the 17,000+ voices in ElevenLabs' library. According to independent tests, v1 Max answered 98.8% of reasoning questions correctly by voice, while v1 Mini got 75.9%. They also respond quickly—v1 Max starts speaking in just 0.57 seconds on average. In tests of conversational dynamics like pauses and interruptions, v1 Mini scored 95.8% and v1 Max 94.1%. In blind human comparisons, v1 Max scored 998 and v1 Mini 988.

The big deal is that you no longer need to be a configuration expert to build a voice agent that sounds natural. ElevenLabs says getting an agent to feel right involves small judgment calls—like how long to wait after a pause, or how to recover when the user interrupts. Eleven Realtime handles those nuances for you. It's also compatible with OpenAI's Realtime API, so developers already using that can switch by changing the endpoint and model name in their code.

The catch is that Eleven Realtime is a developer tool. You'll need some coding knowledge to use it, either through ElevenLabs' platform (ElevenAgents) or directly via the API. It's not a consumer app you can just download and use. But for businesses that want to add voice assistants to their phone lines, websites, or messaging apps, this could make it much faster and cheaper to get started.