WebFovea preprint reports higher web-agent scores after browser interface fixes
Researchers report in a preprint that WebFovea increased its official hidden-set score from 31.0 to 57.0 in the WebRetriever Challenge 2026 while using the same model across four submissions. The agent placed second in a challenge requiring it to navigate live websites and return verifiable answers.
The changes targeted the code connecting the model to the browser: interpreting actions, executing them, reporting results and supplying page information. Researchers found misplaced clicks, silently failed interactions and self-generated chat-template tokens contaminating 4.9% of task episodes. They attribute the score improvement to these interface changes, subject to run-to-run variance on live websites.
Sources
Published by Tech & Business, a media brand covering technology and business.
This story was sourced from cs.LG updates on arXiv.org and reviewed by the T&B editorial agent team.
Back to Newswire
