titleWebArena-Pro: A Heterogeneous, Multimodal, Reproducible Benchmark for Web Agents
authorsImene Kerboua, Fatemeh Pesaran Zadeh, Xing Han Lù, Weijian Qi, Alexander Miller, Junyi Song, Yunjia Tian, Dongjin Kang, Seyeon Choi, Marzia Nouri, Ewen Gueguen, Matteo Boglioni, Fengyuan Liu, Zeyi Liao, Mengqi Yuan, Yue Li, Alexandre Lacoste, Alexandre Drouin, Spandana Gella, Huan Sun, Gunhee Kim, Siva Reddy
date2026-11-13
venuearXiv (preprint, not peer reviewed)
idarXiv:2611.17955
affiliationMcGill NLP Group
upvotes15 (Hugging Face Daily Papers, 2026-11-18)
linksabs pdf hf code
cite
abstractWeb agents powered by large language and vision-language models are increasingly applied to realistic browser work that spans heterogeneous applications, multimodal content, and stateful workflows. However, existing reproducible web-agent benchmarks cover only a small number of web applications drawn from a few software categories, and restrict modality to text and vision. Live benchmarks broaden site coverage but sacrifice reproducibility, since pages and data drift between runs. Moreover, existing benchmarks do not meaningfully evaluate whether agents can understand and use audio and video content embedded within web tasks. To address these gaps, we introduce WebArena-Pro, a benchmark comprising 300 tasks across 20 self-hosted web applications in six domain categories, spanning distinct interface conventions, workflows, and data models. Across the evaluated agents, the best performance is achieved by both Codex with GPT 6.1 Sol and Claude Code with Claude Opus 5.5 at 62.3%, followed by OpenCode with Kimi 3.1 Pro at 60.7%. The strongest GenericAgent configurations, Gemini 3.1 Pro omni and Gemma4 31B, both achieve 30.3%. Among reproducible, human-curated web agent benchmarks, WebArena-Pro provides the broadest application coverage and the most comprehensive multimodal support to date. The benchmark treats audio and video as core observations alongside text and vision, with dedicated actions for extracting information from each. WebArena-Pro runs each task in isolation and supports reproducible, parallel evaluation. Tasks are authored through a dedicated annotator interface, filtered by LLM-assisted triage, and finally validated by humans before release. Project page: https://webarena-pro.github.io

bibtex

@misc{kerboua2026webarena,
  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},
  url={https://arxiv.org/abs/2611.17955}
}