ElevenLabs has launched Eleven Realtime, a pair of native speech-to-speech models designed to power live AI voice conversations. The release includes v1 Mini, optimized for natural, low-cost conversations, and v1 Max, built for demanding tasks like customer support where accuracy and reasoning matter most.
Until now, building a voice agent with ElevenLabs meant stitching together separate pieces: a speech-to-text model, a language model, and a text-to-speech engine. Eleven Realtime folds the whole loop — listening, reasoning, and speaking — into a single end-to-end model, so developers get a voice agent that sounds natural without tuning each component themselves.
The benchmark numbers are strong. The Max variant scores 98.8 percent on the Big Bench Audio benchmark with a response time of about 0.57 seconds, and the company says no model on the Artificial Analysis leaderboard is both faster and more accurate right now. The smaller v1 Mini is cheaper to run and can tell when a speaker has finished talking versus just saying “mm-hmm,” keeping conversations flowing naturally.
Both models are available through the ElevenLabs platform’s ElevenAgents product and an OpenAI-compatible Realtime API, with pricing charged per minute of audio plus token usage. The launch lands as voice AI heats up across the industry — with alternatives from Murf, Play.ht, and Descript — and raises the bar for what a single model can do in a live conversation.
Source: The Decoder
