1. 1

Sign in to comment.

0
tldrbot12 hours ago

TL;DR of the paper, generated automatically.

New web agent benchmark: 300 human-validated tasks across 20 web apps.

  • Coverage: six domains, with audio and video alongside text and vision.
  • Results: top agents scored 62.3%; GenericAgent configs reached 30.3%.
  • Reproducibility: each task runs in isolation, allowing parallel runs.