Squishy

GitHub ↗

A corpus of realistic stuff you might compress. Squishy is a fixed set of real files: novels, source code, server logs, genome reads, weather tables, databases, executables, a photo, a film clip, the weights of a neural network. They range from a few megabytes to a few gigabytes, and every one is freely redistributable, with its source and checksum published. Run any compressor over the set and you get a single Squishy Score you can cite and compare. It's a successor to the Silesia corpus.

The shape of the corpus

Each dot is one file, placed by three measurements of its bytes. No compressor is involved in the placement.

Color is the kind of data. The corpus is chosen so the dots spread across the whole space instead of piling up in one corner.

Drag to rotate · scroll to zoom · hover a dot for details · keyboard: arrows rotate, +/− zoom, 0 resets, Enter steps through.

View the data as a table (no 3D required)
categoryfileentropy (bpb)coveragematch distancesize
Binary & Mediaweights6.240%1 KB269.1 MB
Binary & Mediaexe6.3526%3 KB62.5 MB
Binary & Mediamovie7.980%25 KB12.9 MB
Binary & Mediaphoto7.950%16 B6.5 MB
Binary & Mediawinexe6.2621%1 KB4.1 MB
Binary & Mediaarmexe6.5811%1 KB1.2 MB
Binary & Mediasymbols4.5410%3 KB1.1 MB
Binary & Mediawasm5.682%272 B0.9 MB
Code & Webllvm-project-19.1.0.src.tar5.6968%10 KB1.8 GB
Code & Webclang-releases-16-17-18-19.tar5.6087%21 KB1.5 GB
Code & Webenwik9.txt5.1630%1.1 MB1.0 GB
Code & Webmonorepo5.1534%20 KB50.9 MB
Code & Webmarkup5.2234%14 KB8.0 MB
Code & Webminjs5.479%33 KB3.6 MB
Prosedickens4.652%860 KB12.2 MB
Proseaozora4.8010%1.1 MB12.0 MB
Scale tiernoaa-ghcn-daily-2024-full.csv4.0778%1.5 MB1.3 GB
Scale tierbig-buck-bunny-1080p.mov7.990%53 KB725.1 MB
Structuredecoli-DRR002013-full.fastq3.3167%1 KB1.1 GB
Structurednasa-http-jul-aug-1995.log5.2483%37 KB373.1 MB
Structuredlog5.2273%21 KB26.2 MB
Structuredgenome3.2849%16 B26.2 MB
Structuredjson5.0570%12 KB14.2 MB
Tabular / DBnoaa-ghcn-daily-2021-2023.csv4.0580%2.9 MB4.1 GB
Tabular / DBbts-ontime-2022-20247.498%227 KB772.3 MB
Tabular / DBsqlite5.639%4 KB48.3 MB
Tabular / DBcsv4.0357%152 KB26.5 MB
Tabular / DBparquet7.478%293 KB20.9 MB

The Squishy Score is the geomean of compression across all files - a stable number reference number.

Score your tool

Give it any compressor that reads stdin and writes stdout, such as "xz -9 -c" or your own "./mytool -c". It streams the corpus, runs your tool on every file, and prints the score:

uv run squishy-calculate --cmd "zstd -19 -c"

The Squishy Score is the geometric mean of each file's compression ratio (original size ÷ compressed size). Every file counts once; nothing is weighted, excluded, or tuned. Beside it the runner prints corpus bpb, the plain bits-per-byte over all bytes, where the big files dominate.

The Squishy Score board

Six familiar codecs over the whole corpus — every file counts once. The headline column is the Squishy Score (geomean of per-file ratio); corpus bpb is the byte-weighted companion. Click a column to sort.

toolSquishy Score (×)corpus bpbProseCode & WebStructuredTabular / DBBinary & Media
xz -9 v5.8.35.10×1.0864.15×7.47×12.46×5.48×2.20×
brotli -11 v1.2.05.03×1.1074.11×7.50×12.60×5.24×2.16×
zstd -22 v1.5.74.94×1.1084.10×7.48×13.04×4.96×2.06×
zstd -19 v1.5.74.74×1.1894.07×7.15×11.36×4.89×2.06×
bzip2 -9 4.45×1.3454.02×6.73×11.79×4.01×1.94×
gzip -9 v1.143.63×1.5942.84×5.11×8.20×3.51×1.84×

How to cite

Cite the dated edition you ran against. The DOI resolves to the immutable Zenodo deposit — corpus bytes, score, and metadata — so a citation can never drift from what you measured.

Jack Danger. Squishy-2026: a citable compression benchmark corpus and score. 2026.
DOI: 10.5281/zenodo.XXXXXXX

The machine-readable citation (and BibTeX) live in CITATION.cff. The XXXXXXX is assigned when the edition is minted, then filled in once and never changed.

Every file

What each file is, a peek inside, and how each tool compresses it.

Prose

1

dickens

12.2 MB · Public-Domain

Nine novels by Charles Dickens — English prose.

download dickens · source ↗
compression — 6 tools
bzip2 -9 3.73×
xz -9 v5.8.33.66×
brotli -11 v1.2.03.66×
zstd -22 v1.5.73.63×
zstd -19 v1.5.73.63×
gzip -9 v1.142.69×
2

aozora

12.0 MB · Public-Domain

Collected works of Natsume Sōseki — Japanese literary prose.

download aozora.txt · source ↗
compression — 6 tools
xz -9 v5.8.34.69×
zstd -22 v1.5.74.63×
brotli -11 v1.2.04.62×
zstd -19 v1.5.74.56×
bzip2 -9 4.34×
gzip -9 v1.143.00×

Code & Web

3

monorepo

50.9 MB · Apache-2.0-LLVM

The lib/ source tree of the LLVM Clang C++ compiler.

download monorepo.tar · source ↗
compression — 6 tools
xz -9 v5.8.37.54×
brotli -11 v1.2.07.52×
zstd -22 v1.5.77.47×
zstd -19 v1.5.77.29×
bzip2 -9 6.80×
gzip -9 v1.145.18×
4

minjs

3.6 MB · MIT

The minified Plotly.js charting library — one big line of JavaScript.

download minjs.min.js · source ↗
compression — 6 tools
brotli -11 v1.2.04.29×
xz -9 v5.8.34.26×
zstd -22 v1.5.74.06×
zstd -19 v1.5.74.06×
bzip2 -9 3.81×
gzip -9 v1.143.31×
5

markup

8.0 MB · Freely-distributable

Shakespeare's plays, marked up in XML.

download markup.xml · source ↗
compression — 6 tools
bzip2 -9 5.36×
brotli -11 v1.2.04.87×
xz -9 v5.8.34.83×
zstd -22 v1.5.74.82×
zstd -19 v1.5.74.82×
gzip -9 v1.143.71×

Structured

6

json

14.2 MB · Public-Domain

20,000 magnitude-4.5+ earthquakes, 2010–2024 (USGS GeoJSON).

download data.json · source ↗
compression — 6 tools
bzip2 -9 14.83×
brotli -11 v1.2.013.90×
zstd -22 v1.5.713.23×
zstd -19 v1.5.713.21×
xz -9 v5.8.312.82×
gzip -9 v1.149.43×
7

log

26.2 MB · Public-Domain

A NASA web server's access log from July 1995.

download access.log · source ↗
compression — 6 tools
bzip2 -9 16.97×
brotli -11 v1.2.016.10×
zstd -22 v1.5.715.71×
zstd -19 v1.5.715.56×
xz -9 v5.8.315.00×
gzip -9 v1.1410.36×
8

genome

26.2 MB · Public-Domain

Sequencing reads from an E. coli genome (FASTQ).

download ecoli.fastq · source ↗
compression — 6 tools
brotli -11 v1.2.08.94×
xz -9 v5.8.38.78×
zstd -22 v1.5.78.23×
zstd -19 v1.5.77.52×
bzip2 -9 7.24×
gzip -9 v1.145.97×

Tabular / DB

9

csv

26.5 MB · Public-Domain-USGov

Daily weather observations from NOAA's global climate network, 2024 (CSV).

download data.csv · source ↗
compression — 6 tools
xz -9 v5.8.313.54×
brotli -11 v1.2.013.33×
zstd -22 v1.5.712.45×
zstd -19 v1.5.712.20×
bzip2 -9 9.69×
gzip -9 v1.147.95×
10

parquet

20.9 MB · Public-Domain-USGov

U.S. airline on-time flight records (Bureau of Transportation Statistics) — stored column-wise as Apache Parquet.

download data.parquet · source ↗
compression — 6 tools
brotli -11 v1.2.01.87×
xz -9 v5.8.31.85×
zstd -22 v1.5.71.81×
zstd -19 v1.5.71.81×
bzip2 -9 1.51×
gzip -9 v1.141.48×
11

sqlite

48.3 MB · Public-Domain-USGov

USDA's nutrition database — foods, nutrients, and portions across 17 related tables (SR Legacy).

download data.sqlite · source ↗
compression — 6 tools
xz -9 v5.8.37.56×
brotli -11 v1.2.06.06×
zstd -22 v1.5.75.52×
zstd -19 v1.5.75.51×
bzip2 -9 4.73×
gzip -9 v1.143.75×

Binary & Media

12

exe

62.5 MB · Apache-2.0

A compiled Linux executable — the Hugo static-site generator.

download tool.bin · source ↗
compression — 6 tools
xz -9 v5.8.34.15×
brotli -11 v1.2.04.01×
zstd -22 v1.5.73.78×
zstd -19 v1.5.73.76×
bzip2 -9 3.24×
gzip -9 v1.142.99×
13

photo

6.5 MB · Public-Domain

NASA's “Blue Marble” — Earth photographed from Apollo 17.

download photo.jpg · source ↗
compression — 6 tools
bzip2 -9 1.02×
brotli -11 v1.2.01.02×
zstd -19 v1.5.71.01×
zstd -22 v1.5.71.01×
xz -9 v5.8.31.01×
gzip -9 v1.141.00×
14

movie

12.9 MB · CC-BY-3.0

A clip from the open film Big Buck Bunny (H.264 video).

download movie.mp4 · source ↗
compression — 6 tools
brotli -11 v1.2.01.01×
zstd -19 v1.5.71.01×
zstd -22 v1.5.71.01×
xz -9 v5.8.31.00×
gzip -9 v1.141.00×
bzip2 -9 1.00×
15

weights

269.1 MB · Apache-2.0

The trained weights of a small neural network (safetensors).

download weights.safetensors · source ↗
compression — 6 tools
bzip2 -9 1.44×
xz -9 v5.8.31.42×
brotli -11 v1.2.01.39×
zstd -19 v1.5.71.31×
zstd -22 v1.5.71.30×
gzip -9 v1.141.26×
16

symbols

1.1 MB · MIT

DWARF debug symbols from a Lua 5.4.8 build compiled with -g (a debug-info file, not a runnable program).

download symbols.dwarf · source ↗
compression — 6 tools
xz -9 v5.8.33.78×
brotli -11 v1.2.03.66×
zstd -22 v1.5.73.50×
zstd -19 v1.5.73.50×
bzip2 -9 2.95×
gzip -9 v1.142.90×
17

wasm

868 KB · Public Domain

The SQLite engine compiled to WebAssembly — stack-machine bytecode.

download engine.wasm · source ↗
compression — 6 tools
xz -9 v5.8.32.54×
brotli -11 v1.2.02.49×
zstd -22 v1.5.72.40×
zstd -19 v1.5.72.40×
bzip2 -9 2.35×
gzip -9 v1.142.16×
18

winexe

4.1 MB · MIT OR Apache-2.0

The fd file-finder as a Windows PE executable.

download winexe.exe · source ↗
compression — 6 tools
xz -9 v5.8.33.61×
brotli -11 v1.2.03.52×
zstd -22 v1.5.73.31×
zstd -19 v1.5.73.31×
bzip2 -9 2.80×
gzip -9 v1.142.68×
19

armexe

1.2 MB · MIT

The hyperfine benchmarking tool as an ARM64 Linux executable.

download armexe.elf · source ↗
compression — 6 tools
xz -9 v5.8.32.69×
brotli -11 v1.2.02.56×
zstd -22 v1.5.72.34×
zstd -19 v1.5.72.34×
bzip2 -9 2.15×
gzip -9 v1.142.03×

The big ones

The same kinds of data at gigabyte scale, where long-range matching and big compression windows start to matter. Most are scored cells in the board above; the largest few — the 1080p film, the full-year NOAA CSV, and the two larger Qwen weights — are non-scored diagnostics (throughput and behaviour at extreme sizes, never folded into the Squishy Score).

scored

nasa-http-jul-aug-1995.log

373.1 MB · Public-Domain

Scale-tier file — a large rung for large-window and throughput testing.

download ↓ · sha256 35c38d9465a8ed27… · source ↗
diagnostic

big-buck-bunny-1080p.mov

725.1 MB · CC-BY-3.0

Non-scored diagnostic — distributed for throughput/behaviour at scale, never in the score. The full open film Big Buck Bunny in 1080p H.264 video.

download ↓ · sha256 dc2146a2b1172def… · source ↗
scored

bts-ontime-2022-2024.parquet

772.3 MB · Public-Domain-USGov

Scale-tier file — a large rung for large-window and throughput testing.

download ↓ · sha256 acb6eeb73e9c4449… · source ↗
diagnostic

weights-qwen2.5-0.5b.safetensors

988.1 MB · Apache-2.0

Non-scored diagnostic — distributed for throughput/behaviour at scale, never in the score. Qwen2.5-0.5B — a 0.5B-parameter language model's weights (Apache-2.0). The lower rung of the weights size-ladder above the scored SmolLM2-135M cell.

download ↓ · sha256 fdf756fa7fcbe740… · source ↗
scored

enwik9.txt

1.00 GB · CC-BY-SA-3.0

The first billion bytes of an English Wikipedia XML dump (the Hutter-Prize text).

download ↓ · sha256 159b85351e5f76e6… · source ↗
scored

ecoli-DRR002013-full.fastq

1.07 GB · Public-Domain

Scale-tier file — a large rung for large-window and throughput testing.

download ↓ · sha256 ff3de7024de4f45e… · source ↗
diagnostic

noaa-ghcn-daily-2024-full.csv

1.33 GB · Public-Domain-USGov

Non-scored diagnostic — distributed for throughput/behaviour at scale, never in the score. Scale-tier file — a large rung for large-window and throughput testing.

download ↓ · sha256 70baf8b1fe829889… · source ↗
scored

clang-releases-16-17-18-19.tar

1.50 GB · Apache-2.0-LLVM

Four LLVM/Clang release source trees concatenated — a real software archive.

download ↓ · sha256 e8518848a41185c7… · source ↗
scored

llvm-project-19.1.0.src.tar

1.77 GB · Apache-2.0-LLVM

Scale-tier file — a large rung for large-window and throughput testing.

download ↓ · sha256 bb4ae7add97894e6… · source ↗
diagnostic

weights-qwen2.5-1.5b.safetensors

3.09 GB · Apache-2.0

Non-scored diagnostic — distributed for throughput/behaviour at scale, never in the score. Qwen2.5-1.5B — a larger (1.5B-parameter) language model's weights (Apache-2.0). The top rung of the ladder; multi-GB, for large-window and throughput work.

download ↓ · sha256 dd924a11b4c220f3… · source ↗
scored

noaa-ghcn-daily-2021-2023.csv

4.07 GB · Public-Domain-USGov

Scale-tier file — a large rung for large-window and throughput testing.

download ↓ · sha256 9111537b27d9ed83… · source ↗