Benchmarks
Measured against the tools you know.
15 published results. Each says what was measured. Where Webfold lost, we say so. We say so when the gap closes too.
Jobs
How fast is each job on its own?
The performance page lists each job. It shows what goes in. It shows what comes out. It shows speed. It shows throughput.
Comparisons
Each multiple has its limit.
- 10.8× faster cold start vs Firecrawl Conversion only. See the proof →
- 24× less memory per task vs Headless Chrome One host only. See the proof →
- 1,928 / 1,928 tasks passed on Lexbench Our own run. See the proof →
- 11× more pages a second vs Crawl4AI One host. Directional. See the proof →
- 8.5× more documents a second vs anydoc Cold start gap is 1.1×. See the proof →
- 777× more calls a second vs MarkItDown Start-up is most of it. See the proof →
- 0.023 error rate on a fuzzy scan against 0.628 for pdftotext Eight small files. See the proof →
- 6.0× faster LaTeX to Markdown vs Pandoc Hand-made fixtures. See the proof →
- 8.8× more tasks a second vs Headless Chrome One fixture. One VM. See the proof →
- 275× faster on a script page vs Scrapling Plain fetch: Scrapling wins. See the proof →
- 2.6× faster at search results vs OpenSERP Saved pages only. See the proof →
- 4.8× faster string extraction vs JavaScript parsers Run by the project itself. See the proof →
- 470× faster PII detection vs Presidio without a model Names are far slower. See the proof →
- 26,675 archived pages a second against 2,943 on one thread Eight threads here. See the proof →
All results
Every benchmark has a chart.
Our rules.
A number without a protocol is not published.
Every result names its dataset. It names its sample size. It names its date. It links to the file it came from.
A multiple is two measured values divided. The page shows both. Competitor numbers are theirs. We never use them as neutral evidence. Where we lost or have not measured, we say so.
Some of our earlier figures did not match the retained results. We checked again. We replaced them. We say so on the page.
Run it on your own pages.
Benchmarks are a starting point. Your content is the real test.