New tool2026-11-04via GitHubAI Coding Tools for Developers

WebsiteBench

Name: WebsiteBench
Author: TIGER-AI-Lab

WebsiteBench is an open-source benchmark from TIGER-AI-Lab that asks AI coding agents to explore 70 real production websites through a browser and rebuild each one as a full-stack app that looks and behaves the same, scored against 46,916 hidden test cases.

Visit WebsiteBench ↗Repository ↗

Free (open source)Apache-2.0 (code, tasks and verifier); reference sites are for non-commercial research onlyPython (needs Docker and the Harbor CLI)★ 151 (as of 2026-11-09)by TIGER-AI-Lab

Why it matters

For teams choosing or building coding agents, it measures something closer to real work than small coding puzzles: whether an agent can reverse-engineer a whole working website just by clicking around it.

Highlights

70 website tasks with 302 to 2,016 hidden test cases each, 46,916 in total, published on Harbor Hub and Hugging Face.
A case passes only when its functional checks pass and every visual checkpoint matches the original closely (a normalized SSIM score of 0.99).
Runs through the Harbor evaluation framework in Docker, for example with Codex and a model of your choice, on all 70 tasks or a single one.
A websitebench package on PyPI downloads the tasks, and each task ships a reference solution you can run to check the setup.
A public leaderboard of results is on the project site.unverified

Taken from the project's own README, docs or launch page. Claims about speed or accuracy are the maker's unless stated otherwise.

Try it if: You're comparing coding agents or models and want a hard, realistic test of full-stack web building.

Worth knowing: The reference websites copy third-party sites and are licensed for non-commercial research only. The README lists a paper badge but no paper link yet.

Quick facts

Pricing	Free; running agents costs whatever your model API usage costs
Free or paid	Free (open source)
License	Apache-2.0 (code, tasks and verifier); reference sites are for non-commercial research only
Language	Python (needs Docker and the Harbor CLI)
Platform	Any machine that runs Docker; cloud runs documented
GitHub stars	★ 151 (as of 2026-11-09)
Maker	TIGER-AI-Lab

Related ranking

See how established tools compare in Best AI Coding Tools for Developers. New releases aren't scored until we review them against the rubric.

What we couldn't verify

highlights[4] (leaderboard contents not reviewed)

Checked 2026-11-09 from the tool's own pages. Prices and star counts change, so confirm on the official site. Listing here is not a ranking or an endorsement.