AI & Tech··1 sources verified
EverydayBench Finds Top AI Agents Complete Only 54.55% of Live Web Tasks
Summarised by Relevant News AI · Read time: 3 min
Researchers introduced EverydayBench, a benchmark of 77 human-written tasks across eight everyday life and work domains, which evaluates agents on live web outcomes. It scores submissions with deterministic constraint checks, verification of claims against cited webpages retrieved at evaluation time, and rubric-based grading of deliverable quality. Across 23 agent-model configurations, Claude Opus 5.5 and Claude Fable 5.1 tied for the highest strict pass rate at 54.55%, and reliability varied by task type.
Why it matters: Teams building or buying web-browsing agents get a more rigorous way to measure whether agents deliver complete, well-sourced, usable results rather than just retrieving information.
All sources