# Vision2Web: A Hierarchical Benchmark for Visual Website Development with Agent Verification

> Benchmarks website development from visual prototypes and written requirements across 100 static-page, 66 interactive-frontend and 27 full-stack tasks, combining workflow-based functional checks with model-judged visual comparison.

- Authors: Zehai He; Wenyi Hong; Zhen Yang; Ziyang Pan; Mingdao Liu; Xiaotao Gu; Jie Tang
- Venue: ICML 2026
- Status: Conference paper
- Publication / posting date: September 29, 2026
- Date source: Publication date stated for PMLR volume 306; the July 6–11 conference dates are not used as the publication date.
- First arXiv submission: March 27, 2026
- arXiv: 2603.26648
- Reviewed arXiv version: v3
- Sources reviewed: 2026-10-17
- Added to the catalog: 2026-10-17
- Original paper: <https://proceedings.mlr.press/v306/he26n.html>
- PDF: <https://raw.githubusercontent.com/mlresearch/v306/main/assets/he26n/he26n.pdf>
- Code: <https://github.com/zai-org/Vision2Web>
- Project: <https://vision2web-bench.github.io/>
- BibTeX: <https://awesomerea.org/papers/citations/paper-vision2web.bib>
- Catalog page: <https://awesomerea.org/papers/paper-vision2web/>
- Record JSON: <https://awesomerea.org/papers/paper-vision2web/record.json>
- Chinese version: [简体中文](https://awesomerea.org/zh/papers/paper-vision2web/index.md)

## Scope and limitations

The full-stack tier is specification-driven, not reconstruction solely by exploring a live website. Workflows test state management and CRUD outcomes, but finite agent judgments do not establish exact backend or durable-database equivalence. The current public evaluator maintains a browser session within each workflow and permits local business-API checks; this is not a universal persistence proof.

## Why it fits this collection

A useful comparison for evaluating interactive and stateful website reconstruction, while separating written-specification development from behavior-only inference.

## Publication and review notes

The final PMLR paper has seven authors and pages 41393–41415; arXiv v3 is dated 2026-07-20. Current artifact evaluation differs from the paper configuration. Licensing signals conflict: README text says CC-BY-NC-SA-4.0, Hugging Face metadata says Apache-2.0, the project page says CC BY-SA 4.0, and setup.py lists MIT while the repository has no LICENSE file. No uniform artifact license is asserted.

## Primary sources

1. <https://proceedings.mlr.press/v306/he26n.html>
2. <https://proceedings.mlr.press/v306/>
3. <https://raw.githubusercontent.com/mlresearch/v306/main/assets/he26n/he26n.pdf>
4. <https://arxiv.org/abs/2603.26648>
5. <https://arxiv.org/html/2603.26648v3>
6. <https://vision2web-bench.github.io/>
7. <https://github.com/zai-org/Vision2Web/blob/d80a4a3d0ea7326541ad87314294269a633789e4/README.md>
8. <https://github.com/zai-org/Vision2Web/blob/d80a4a3d0ea7326541ad87314294269a633789e4/setup.py>
9. <https://github.com/zai-org/Vision2Web/blob/d80a4a3d0ea7326541ad87314294269a633789e4/vision2web/evaluation/prompts.py>
10. <https://github.com/zai-org/Vision2Web/blob/d80a4a3d0ea7326541ad87314294269a633789e4/vision2web/evaluation/functional_tester.py>
11. <https://huggingface.co/datasets/zai-org/Vision2Web/blob/8f03299d92b9bd852e93852d0c21e8a4848ab661/README.md>

## About this entry

Reviewed against the paper and its primary sources. Cite the original authors using the linked paper or BibTeX.
Original editorial content: MIT. Linked projects, papers and assets retain their own rights and licenses.