Mind2Web 2: Evaluating Agentic Search with Agent-as-a-Judge

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Gou, Boyu, Huang, Zanming, Ning, Yuting, Gu, Yu, Lin, Michael, Qi, Weijian, Kopanev, Andrei, Yu, Botao, Gutiérrez, Bernal Jiménez, Shu, Yiheng, Song, Chan Hee, Wu, Jiaman, Chen, Shijie, Moussa, Hanane Nour, Zhang, Tianshu, Xie, Jian, Li, Yifei, Xue, Tianci, Liao, Zeyi, Zhang, Kai, Zheng, Boyuan, Cai, Zhaowei, Rozgic, Viktor, Ziyadi, Morteza, Sun, Huan, Su, Yu
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866909673979904000
author Gou, Boyu
Huang, Zanming
Ning, Yuting
Gu, Yu
Lin, Michael
Qi, Weijian
Kopanev, Andrei
Yu, Botao
Gutiérrez, Bernal Jiménez
Shu, Yiheng
Song, Chan Hee
Wu, Jiaman
Chen, Shijie
Moussa, Hanane Nour
Zhang, Tianshu
Xie, Jian
Li, Yifei
Xue, Tianci
Liao, Zeyi
Zhang, Kai
Zheng, Boyuan
Cai, Zhaowei
Rozgic, Viktor
Ziyadi, Morteza
Sun, Huan
Su, Yu
author_facet Gou, Boyu
Huang, Zanming
Ning, Yuting
Gu, Yu
Lin, Michael
Qi, Weijian
Kopanev, Andrei
Yu, Botao
Gutiérrez, Bernal Jiménez
Shu, Yiheng
Song, Chan Hee
Wu, Jiaman
Chen, Shijie
Moussa, Hanane Nour
Zhang, Tianshu
Xie, Jian
Li, Yifei
Xue, Tianci
Liao, Zeyi
Zhang, Kai
Zheng, Boyuan
Cai, Zhaowei
Rozgic, Viktor
Ziyadi, Morteza
Sun, Huan
Su, Yu
contents Agentic search such as Deep Research systems-where agents autonomously browse the web, synthesize information, and return comprehensive citation-backed answers-represents a major shift in how users interact with web-scale information. While promising greater efficiency and cognitive offloading, the growing complexity and open-endedness of agentic search have outpaced existing evaluation benchmarks and methodologies, which largely assume short search horizons and static answers. In this paper, we introduce Mind2Web 2, a benchmark of 130 realistic, high-quality, and long-horizon tasks that require real-time web browsing and extensive information synthesis, constructed with over 1000 hours of human labor. To address the challenge of evaluating time-varying and complex answers, we propose a novel Agent-as-a-Judge framework. Our method constructs task-specific judge agents based on a tree-structured rubric design to automatically assess both answer correctness and source attribution. We conduct a comprehensive evaluation of ten frontier agentic search systems and human performance, along with a detailed error analysis to draw insights for future development. The best-performing system, OpenAI Deep Research, can already achieve 50-70% of human performance while spending half the time, highlighting its great potential. Altogether, Mind2Web 2 provides a rigorous foundation for developing and benchmarking the next generation of agentic search systems.
format Preprint
id arxiv_https___arxiv_org_abs_2506_21506
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Mind2Web 2: Evaluating Agentic Search with Agent-as-a-Judge
Gou, Boyu
Huang, Zanming
Ning, Yuting
Gu, Yu
Lin, Michael
Qi, Weijian
Kopanev, Andrei
Yu, Botao
Gutiérrez, Bernal Jiménez
Shu, Yiheng
Song, Chan Hee
Wu, Jiaman
Chen, Shijie
Moussa, Hanane Nour
Zhang, Tianshu
Xie, Jian
Li, Yifei
Xue, Tianci
Liao, Zeyi
Zhang, Kai
Zheng, Boyuan
Cai, Zhaowei
Rozgic, Viktor
Ziyadi, Morteza
Sun, Huan
Su, Yu
Artificial Intelligence
Computation and Language
Agentic search such as Deep Research systems-where agents autonomously browse the web, synthesize information, and return comprehensive citation-backed answers-represents a major shift in how users interact with web-scale information. While promising greater efficiency and cognitive offloading, the growing complexity and open-endedness of agentic search have outpaced existing evaluation benchmarks and methodologies, which largely assume short search horizons and static answers. In this paper, we introduce Mind2Web 2, a benchmark of 130 realistic, high-quality, and long-horizon tasks that require real-time web browsing and extensive information synthesis, constructed with over 1000 hours of human labor. To address the challenge of evaluating time-varying and complex answers, we propose a novel Agent-as-a-Judge framework. Our method constructs task-specific judge agents based on a tree-structured rubric design to automatically assess both answer correctness and source attribution. We conduct a comprehensive evaluation of ten frontier agentic search systems and human performance, along with a detailed error analysis to draw insights for future development. The best-performing system, OpenAI Deep Research, can already achieve 50-70% of human performance while spending half the time, highlighting its great potential. Altogether, Mind2Web 2 provides a rigorous foundation for developing and benchmarking the next generation of agentic search systems.
title Mind2Web 2: Evaluating Agentic Search with Agent-as-a-Judge
topic Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2506.21506