🌐 WebRetriever: A Large-Scale Comprehensive Benchmark for Efficient Web Agent Evaluation

ECCV 2026

📃 Paper • 🏆 Leaderboard • 💻 Code

---

Overview

Existing web agent evaluation suffers from three key limitations: (1) insufficient benchmark coverage — offline benchmarks lack real-world fidelity, while online benchmarks remain limited in website scale, domain diversity, and intent variety, leading to biased and overly optimistic assessments; (2) unscalable evaluation methods — human annotation is costly and impractical at scale, and current automated approaches lack sufficient accuracy on complex queries; (3) narrow evaluation dimensions — existing protocols fail to systematically assess agents' ability to leverage external knowledge and execute end-to-end workflows required for real-world deployment. WebRetriever addresses all three gaps through large-scale diverse benchmarking, a reliable automated evaluation method, and deployment-oriented evaluation protocols.

Dataset

Comparison with Existing Benchmarks

Intent-Type	Benchmarks	Setting	Online	Interactive	Websites	Tasks
General	MiniWoB	Synthetic	❌	✔️	100	100
	MiniWoB++	Synthetic	❌	✔️	100	100
	WebArena	Semi-real	❌	✔️	4	812
	VisualWebArena	Semi-real	❌	✔️	3	910
	REAL	Semi-real	❌	✔️	11	112
	WebLINX	Real	❌	❌	155	1,368
	Mind2Web	Real	❌	❌	137	1,341
	Mind2Web-Live	Real	✔️	✔️	46	104
	MMInA	Real	✔️	✔️	14	1,050
	AssistantBench	Real	✔️	✔️	258	214
	WebVoyager	Real	✔️	✔️	15	643
	Bearcubs	Real	✔️	✔️	108	111
	Online-Mind2Web	Real	✔️	✔️	136	300
Professional	WebShop	Semi-real	❌	✔️	1	500
	ST-WebAgentBench	Semi-real	❌	✔️	3	222
	Wonderbread	Semi-real	❌	✔️	4	598
	WorkArena	Real	✔️	✔️	5	33
	WorkArena++	Real	✔️	✔️	5	682
	OmniACT	Real	✔️	✔️	27	736
General & Professional	WebRetriever	Real	✔️	✔️	800	1,550

Three Evaluation Protocols (1,550 tasks)

Protocol	Description
Protocol I	Navigation-only: reach the target page based on task description (1,000 tasks)
Protocol II	Navigation with operational documentation as guidance (1,000 tasks)
Protocol III	End-to-end: navigation + information extraction (100 tasks)

Notably, Protocols I and II share 550 overlapping tasks, with the sole difference being the availability of operational documentation.

Citation

@misc{dong2026webretrieverlargescalecomprehensivebenchmark,
  title={WebRetriever: A Large-Scale Comprehensive Benchmark for Efficient Web Agent Evaluation}, 
  author={Wei Dong and Tianyu Fu and Zhe Yu and Hanning Wang and Anyang Su and Zhizhou Fang and Yuyang Chen and Shuo Wang and Minghui Wu and Ping Jiang and Zhen Lei and Chenxu Zhao},
  year={2026},
  eprint={2607.06118},
  archivePrefix={arXiv},
  primaryClass={cs.CV},
  url={https://arxiv.org/abs/2607.06118}, 
}

License

This dataset is released under the CC BY 4.0 license.