WebArena-Pro
A Heterogeneous, Multimodal, Reproducible
Benchmark for Web Agents
Latest news
- — WebArena-Pro will be presented at NeurIPS Paris.
- — WebArena-Pro is released! 🎉 🎉
- — WebArena-Pro has been accepted at NeurIPS 2026 · E&D Track.
What is WebArena-Pro?
WebArena-Pro evaluates AI web agents on 300 tasks across 20 self-hosted applications.
Tasks span six domains, from productivity and communication to analytics, business, media, and travel.
Agents read text, interpret images, listen to audio, and watch video to complete their tasks.
Each task runs in an isolated environment, so evaluations can be repeated and compared.
Task examples
Task examples from 19 benchmark websites. Choose an application to read its task and inspect a screenshot of the seeded environment.
Leaderboard
Submit resultsScroll horizontally to see all columns.
| Rank | Agent | Model | Success requires a final reward of 1. | Traces | Paper | Code |
|---|
Select 50 steps to explore the reported results.
Scroll horizontally to see all domains.
| Agent | Model | Productivity | Communication | Analytics | Business operations |
Design & media |
Travel | Success requires a final reward of 1. |
|---|
Select 50 steps to explore the reported results.
Paper and citation
Use the citation below to reference WebArena-Pro.
@misc{kerboua2026webarenaproheterogeneousmultimodalreproducible,
title={WebArena-Pro: A Heterogeneous, Multimodal, Reproducible Benchmark for Web Agents},
author={Imene Kerboua and Fatemeh Pesaran Zadeh and Xing Han Lù and Weijian Qi
and Alexander Miller and Junyi Song and Yunjia Tian and Dongjin Kang
and Seyeon Choi and Marzia Nouri and Ewen Gueguen and Matteo Boglioni
and Fengyuan Liu and Zeyi Liao and Mengqi Yuan and Yue Li
and Alexandre Lacoste and Alexandre Drouin and Spandana Gella
and Huan Sun and Gunhee Kim and Siva Reddy},
year={2026},
eprint={2611.17955},
archivePrefix={arXiv},
primaryClass={cs.AI},
url={https://arxiv.org/abs/2611.17955},
}