Skip to content

docs: add the /vs/ comparison hub and anymd vs Kreuzberg and pdftotext pages - #1033

Merged
shtse8 merged 1 commit into
mainfrom
feat/vs-hub-kreuzberg-pdftotext
Nov 18, 2026
Merged

shtse8 merged 1 commit into
mainfrom
feat/vs-hub-kreuzberg-pdftotext

Conversation

@shtse8

@shtse8 shtse8 commented Nov 18, 2026

Copy link
Copy Markdown
Collaborator

https://sylphxai.github.io/anymd/vs/ returned 404, and Kreuzberg and pdftotext had AgentDocBench results but no comparison page.

What ships

  • docs/vs/index.md: all seven tools side by side. The scoreboard (score, documents converted, table cells F1, time, output tokens, version and run date, plus speed and tokens on the 19 documents every tool converted) is generated by bench/leaderboard.py from bench/results/*.json and links every head-to-head page. A short note per tool says when it fits, then the anymd Pro (US$29, one payment) section links to /pricing?entry_surface=seo_vs-hub.
  • docs/vs/kreuzberg.md: 96.4 vs 81.7, 11.5 s vs 16.0 s, 19% fewer tokens, table cells F1 92.2 vs 38.4. The page credits Kreuzberg's lead on HTML and EPUB, its tie on spreadsheets, its bindings and extra formats, and notes that v4 is now a long-term-support line whose successor, Xberg, was not measured.
  • docs/vs/pdftotext.md: 96.4 vs 42.2. pdftotext reads only the 23 PDFs and skips the 15 other files. On those 23 it is faster (0.90 s vs 11.4 s) because anymd spends 10.5 s on OCR for three scans. On the other 20 PDFs the times are 0.87 s and 0.88 s. The page also says that pdftotext's output is the text F1 reference for born-digital PDFs.

Generator

  • When the other tool is faster on the documents both converted (only pdftotext today), the head-to-head block shows how much of anymd's time is OCR on scans and compares the rest. Existing pages are unchanged.
  • pdftotext's version shows as 24.02.0 instead of version 24.02.0.
  • The hub's headline figures and the new pages' figures, including those in llms.txt, are checked against the results by python bench/leaderboard.py --check.

Wiring

The nav "Compare" menu, the sidebar and the footer now start with "All comparisons" (/vs/) and list Kreuzberg and pdftotext. The benchmark guide, the README and llms.txt link them too. The comparison JSON-LD gives the hub …/vs/ as its URL, not …/vs/index. The sitemap is generated by VitePress and picks up the three pages. /pricing links still go through the full-page-load route hook from #1029, and the observability setup from #1032 is untouched.

Tests

test/vs-pages.test.ts now requires a page for every tool in bench/results/ and adds hub checks: nav, sidebar, footer, guide and llms.txt links; a generated scoreboard that links every page; the canonical and JSON-LD URL; the FAQ JSON-LD matching the visible FAQ; Pro attribution; no internal names.

…t pages

/anymd/vs/ returned 404, and Kreuzberg and pdftotext had benchmark results
but no comparison page. The hub shows all seven AgentDocBench tools side by
side, with a scoreboard that bench/leaderboard.py generates from
bench/results/*.json and links to every head-to-head page. The two new pages
follow the Docling, Marker and Unstructured pages, with generated
head-to-head blocks.

When the other tool is faster on the documents both converted, as pdftotext
is, the generated block now shows how much of anymd's time is OCR on scanned
documents and compares the rest. pdftotext's version reads "24.02.0" rather
than "version 24.02.0". The hub's and the new pages' headline figures are
checked against the results like the existing pages.

The nav, sidebar, footer, benchmark guide, README and llms.txt link the hub
and both pages, and the comparison JSON-LD names /vs/ as the hub's URL.
@shtse8
shtse8 enabled auto-merge November 18, 2026 15:06
@shtse8
shtse8 added this pull request to the merge queue Nov 18, 2026
Merged via the queue into main with commit 98f022b Nov 18, 2026
24 checks passed
@shtse8
shtse8 deleted the feat/vs-hub-kreuzberg-pdftotext branch November 18, 2026 15:18
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant