The website of the LeanLean benchmark: https://leanleanbench.com/
It shows the leaderboard and, for every model and repository, what the agent changed: a dependency graph, a code diff, and the agent's full trace.
python3 serve.py 8000Then open http://localhost:8000. The site is plain HTML, CSS and JavaScript in site/,
with its data already built in site/data/.
site/paper.html is a static abstract and citation page next to site/paper.pdf.
It loads no JavaScript, benchmark data, or external assets. Update the title, authors,
year, abstract, and citation URL in pipeline/paper.json, then run:
python3 pipeline/build_paper.py
python3 pipeline/stamp_assets.pyThis also updates the paper metadata used by the benchmark. The full build runs both steps, and CI checks that the generated files are current.
Every Pages deployment regenerates site/assets/og.png from the current frontier
chart, including the published reasoning variants. It also updates the image alt
text and versions the image URL so new previews can fetch the latest card.
To regenerate locally:
uv run --with playwright==1.63.0 playwright install firefox
uv run --with playwright==1.63.0 python pipeline/build_og.pyBefore publishing refreshed data, run python3 pipeline/anonymize_paths.py to
replace usernames in local home-directory paths with user. It updates compressed
trace packs and their byte-range indexes together. Run it with --check to verify
that no further replacements are needed.
- Code (the website): MIT License, see
LICENSE. - Data (
site/data/: results, leaderboard and agent traces): CC BY 4.0, except the third-party content below. - Benchmark repositories: their code shown on the site, and the agents' edits to it, keep their own licenses; see each upstream repository, linked from the site.
- System prompts: the harnesses' built-in system prompts
(
site/data/system-prompts.json) belong to their makers and are reproduced for research. - Logos: trademarks of their owners; see
site/assets/logos/README.md.