Open data
Open datasets
Reproducible benchmarks and computed reference tables from Cleanor Labs, free to reuse under CC BY 4.0. Every dataset is generated by a script you can re-run, published on GitHub, and mirrored to Kaggle and Hugging Face; the compression benchmarks are archived on Zenodo with a citable DOI. This page is the index — the CSVs live in the repositories.
- ~8.9xthe bitrate Xvid/DivX needs over AV1 at the same quality (4x over H.264)
Video codec size benchmark (Xvid/DivX vs H.264/HEVC/AV1/VP9)
Bitrate each codec needs at matched perceptual quality (SSIM), measured on real 720p clips. The data behind "which video codec is smallest".
- 936 MBTikTok; the median popular app is ~400 MB, 9 of 40 exceed 500 MB
App size index: App Store download sizes of popular apps
The measured App Store download size of 40 popular iOS apps, date-stamped. The measured floor behind "why does this app take up so much storage".
- 44–91xhow much smaller the same clip is as an MP4 than as a GIF
Animated format size benchmark (GIF vs APNG vs MP4 vs WebM)
The same short clip encoded as GIF, APNG, MP4 and WebM. Answers "why is my GIF so huge, and how much smaller is an MP4".
- ~10% / ~6%lossless size saved on PNG (oxipng) and baseline JPEG (jpegtran)
Lossless PNG/JPEG recompression savings (oxipng, jpegtran)
How much smaller a PNG or JPEG gets with zero quality loss, verified bit-identical. Better entropy coding, stripped metadata.
- ~11%a TTF as a Latin-subset WOFF2 (WOFF2 alone is ~40% of the TTF)
Font format weight: TTF vs WOFF vs WOFF2 and subsetting
How much a web font shrinks as WOFF2 and when subset to Latin, losslessly, across five popular families.
- 5–16%SVGO on already-clean SVGs; gzip is the bigger lever (~40–50% of original)
SVG minification savings (SVGO + gzip)
How much SVGO shrinks an SVG, and how much more gzip does, across clean icons and detailed emoji.
- ~70% / ~13%gs /ebook (lossy, text kept) vs qpdf (lossless, text byte-identical)
PDF compression benchmark (Ghostscript vs qpdf)
How much a PDF shrinks lossily (Ghostscript image downsampling) vs losslessly (qpdf), with a text-preservation check.
- ~13xmarkup: Apple charges ~$0.78/GB at the base tier vs ~$0.06/GB flash
The storage upgrade tax: device $/GB vs flash cost
What device makers charge to add storage, per extra GB, vs what NAND flash costs. US list prices, date-stamped.
- 10 CSVsexact, reproducible reference tables, computed from the spec
Developer reference datasets (aspect ratios, WCAG contrast, MIME, encoding, QR, format support)
Computed lookup tables developers reach for: aspect-ratio pixels, WCAG 2.1 contrast with AA/AAA, file types with MIME + magic numbers, UTF-8/16/32 sizes, QR capacity, and browser format support.
Prefer a human-readable write-up? See the studies on Research. Every dataset above is CC BY 4.0 — cite Cleanor Labs and reuse it freely.