ElevenLabs Scribe v3 Live uses conversational context to stop voice agents from cutting off callers mid-sentence

Why it matters

ElevenLabs has released two speech-to-text models, Scribe v3 for pre-recorded audio and Scribe v3 Live for real-time use, both designed to handle difficult audio conditions such as crosstalk and background noise. Independent benchmarks from Artificial Analysis place Scribe v3 among the most accurate transcription models available, ahead of competitors from Microsoft, xAI, Google, OpenAI, and Mistral, though it is slower and more expensive than offerings from Microsoft and xAI. The benchmarks measure only word errors on recorded files, leaving claims about speaker diarization, rare languages, and real-time behavior untested. Scribe v3 Live addresses a common frustration with voice agents that cut off callers during brief pauses by analyzing conversational context to judge when a speaker has finished. Developers can preload specialized vocabulary, set formatting rules, and use a multilingual mode. The company is also deepening enterprise ties with Nvidia and McKinsey, and launching a marketplace with vendors such as Salesforce, IBM, and Zendesk.