ElevenLabs ships Eleven v4 Turbo with ~100ms voice latency
The new voice model generates speech at a median latency of about 100ms, below the roughly 200ms rhythm of human conversation, and Klarna and Revolut already use it at scale for phone support.
Human conversation runs on a 200 millisecond rhythm, and ElevenLabs just shipped a voice model that answers in 100. That number is the whole story of this launch. Linguists have measured the gap between one person finishing a sentence and the other person starting across languages all over the world, and it averages around 200 milliseconds everywhere. Your brain is so tuned to this rhythm that anything past half a second starts to register as hesitation or evasion. Voice bots always failed right here. The voices got convincing years ago, but the machine needed a full second to transcribe you, think, and speak, and that beat of dead air told you exactly what you were talking to. Eleven v4 Turbo generates speech at a median of ~100ms. Pair that with fast speech recognition and a fast LLM and the machine answers inside the human rhythm. Sometimes faster than a person who was actually listening to you. And this is already running at scale. Klarna uses an ElevenLabs agent as first-line phone support for 35 million US customers. Revolut runs one across Europe in 30+ languages. The money followed the latency curve. ElevenLabs did $4.6M in revenue in 2023. Sacra estimates they crossed $600M ARR this June, and Bloomberg reported a secondary in the works at a $22B valuation, double what Sequoia paid in February. Humans take 200 milliseconds to answer each other. The machine takes 100. From here on, the pause won't tell you anything.
This post quotes another post: “ElevenLabs releases Eleven v4 and v4 Turbo voice models”