Elevenlabs Offers Two AI Voice Models with Reduced Latency

Elevenlabs launches two real-time speech-to-speech conversion models. The Max version achieves 98.8% on Big Bench Audio with a latency of 0.57 seconds, while the Mini targets more economical use cases. Both are accessible via ElevenAgents and an OpenAI-compatible API, with billing based on minutes and tokens.

Immediate Access via ElevenAgents and OpenAI-Compatible API

The two variants, Eleven Realtime v1 Max and v1 Mini, are available on the ElevenAgents platform as well as through a Realtime API compatible with OpenAI. Pricing is based on usage time per minute, with additional token costs. The v1 Max model is offered at a higher rate than the v1 Mini. Detailed pricing information is available on the Elevenlabs pricing page.

Announced Scores and Position on the Artificial Analysis Ranking

The Eleven Realtime v1 Max version boasts a score of 98.8% on the Big Bench Audio benchmark, with a response time of 0.57 seconds. On the Artificial Analysis Speech to Speech ranking, v1 Max is positioned near the upper left corner, indicating a combination of high accuracy and low latency. According to Elevenlabs, no other model in the ranking is currently both faster and more accurate.

Two Usage Profiles: Natural at Lower Cost and Complex Tasks

The Eleven Realtime v1 Mini model is designed for natural conversations at a reduced cost and can detect the end of speech or acknowledgments like "mm-hmm." The v1 Max is intended for more complex tasks, such as customer support. Both models, Eleven Realtime v1 Max and v1 Mini, are speech-to-speech conversion systems for real-time interactions.