Preprint · 2025
BrowseComp: A Simple Yet Challenging Benchmark for Browsing Agents
| Venue | arXiv |
|---|---|
| DOI | 10.48550/arXiv.2504.12516 |
| URL | http://arxiv.org/abs/2504.12516 |
| Topics | ReplicHAI / Agent Evals / 04 Web browsing & deep research |
| Tags | Computer Science - Computation and Language |
| archiveID | arXiv:2504.12516 |
| extra | arXiv:2504.12516 [cs.CL] |
| libraryCatalog | arXiv.org |
| shortTitle | BrowseComp |
Abstract
We present BrowseComp, a simple yet challenging benchmark for measuring the ability for agents to browse the web. BrowseComp comprises 1,266 questions that require persistently navigating the internet in search of hard-to-find, entangled information. Despite the difficulty of the questions, BrowseComp is simple and easy-to-use, as predicted answers are short and easily verifiable against reference answers. BrowseComp for browsing agents can be seen as analogous to how programming competitions are an incomplete but useful benchmark for coding agents. While BrowseComp sidesteps challenges of a true user query distribution, like generating long answers or resolving ambiguity, it measures the important core capability of exercising persistence and creativity in finding information. BrowseComp can be found at https://github.com/openai/simple-evals.
BibTeX
@misc{wei_browsecomp_2025,
title = {{BrowseComp}: {A} {Simple} {Yet} {Challenging} {Benchmark} for {Browsing} {Agents}},
shorttitle = {{BrowseComp}},
url = {http://arxiv.org/abs/2504.12516},
doi = {10.48550/arXiv.2504.12516},
abstract = {We present BrowseComp, a simple yet challenging benchmark for measuring the ability for agents to browse the web. BrowseComp comprises 1,266 questions that require persistently navigating the internet in search of hard-to-find, entangled information. Despite the difficulty of the questions, BrowseComp is simple and easy-to-use, as predicted answers are short and easily verifiable against reference answers. BrowseComp for browsing agents can be seen as analogous to how programming competitions are an incomplete but useful benchmark for coding agents. While BrowseComp sidesteps challenges of a true user query distribution, like generating long answers or resolving ambiguity, it measures the important core capability of exercising persistence and creativity in finding information. BrowseComp can be found at https://github.com/openai/simple-evals.},
urldate = {2026-11-06},
publisher = {arXiv},
author = {Wei, Jason and Sun, Zhiqing and Papay, Spencer and McKinney, Scott and Han, Jeffrey and Fulford, Isa and Chung, Hyung Won and Passos, Alex Tachard and Fedus, William and Glaese, Amelia},
month = apr,
year = {2025},
note = {arXiv:2504.12516 [cs.CL]},
keywords = {Computer Science - Computation and Language},
}