EvoBrowseComp: Benchmarking Search Agents on Evolving Knowledge

In three linesEvoBrowseComp is an evolving benchmark of 400 English and 400 Chinese questions to evaluate search agents (LLM + web tools). Unlike static BrowseComp, it uses live-web traversal and a three-agent framework (QA synthesis, information filtering, high-level guidance) to prevent contamination and parametric memorization. The benchmark auto-updates regularly.

Summary generated by Claude — human-verified