DailyReport: An Open-ended Benchmark for Evaluating Search Agents on Daily Search Tasks

In three linesDailyReport is an open-source benchmark evaluating search agents on 150 real-world daily tasks with 3,546 evaluation rubrics. Tasks decomposed into subtasks with cascade evaluation across disentangled dimensions. Testing 17 agentic systems reveals significant gaps versus user expectations.

Summary generated by Claude — human-verified