Megjelent az Eleven Realtime beszédből beszéd hangmodell

Introducing Eleven Realtime, a new class of Speech to Speech models. Voice agents tend to fail callers in one of two ways: they pause too long, or they answer quickly and get it wrong. In independent testing by Artificial Analysis, Eleven Realtime v1 Max starts speaking in 0.57 seconds and answers 98.8% of spoken reasoning questions correctly. No model on their leaderboard is both faster and more accurate. A natural-sounding agent depends on small interactions: how long to wait after a pause, how to recover when a caller interrupts, and when “mm-hm” means keep going. Eleven Realtime does that tuning for you, packaging a complete voice agent, optimized end-to-end. There are two models you can choose from. v1 Mini is for conversations that need to feel natural, like AI assistants, tutors and interactive characters. v1 Max is for calls that need to get something done, like customer support, lead qualification, and multi-step requests where the agent looks things up and takes action. Eleven Realtime is now available in ElevenAgents and through the Realtime API.