OpenDS Benchmark for Real-World Web Deep Search Agents
October 18, 2026
The OpenDS task introduces a framework for evaluating agents on multi-step research within live, heterogeneous web environments. It moves beyond static retrieval by requiring autonomous browser interaction, evidence discovery, and integration across diverse web modalities.
HOW THIS AFFECTS YOU
●
builderYou can use this benchmark to test how well your agents handle non-static, real-world web navigation.
●
researcherThis provides a more rigorous evaluation metric for autonomous web-based agents compared to static datasets.