Public test files · cmz 1.0 beta 5

Test files anyone can download

Many of the files behind our main page are private, so nobody else can repeat those measurements. These 27 can be repeated. Each one is published by someone else, anyone can download it, and its SHA-256 checksum shows that you have exactly the bytes we measured. We don't host the files: the script below fetches each one from its publisher and checks it.

64.5 MB27 files, original

30.0 MBafter cmz, 46.5% of the original

19 min 43 sto pack all of them

7 min 0 sto unpack, 27 of 27 identical

Script for Windows (PowerShell) Script for Linux and macOS The list as a table (TSV)

Same 27 files · one PC · lower left is better

Size against time

Every program packed all 27 files with the settings shown. Each point is the total size against the total time.

cmz beta 5: 46.5% in 20 min. It is smaller than paq8px (50.8%) and about 11 times faster. paq8pxd in its slow mode packs tighter (44.1%) and takes 8 times as long.

paq8px and paq8pxd are research compressors, the strongest known family of this kind. They run from the command line on one file at a time. To save a day, paq8pxd -s8 and paq8px ran two at a time; their times here are corrected by the measured slowdown, about 18%.

7-Zip, xz and zpaq are everyday archivers: done in seconds or minutes, but they leave 64–71% of the original, because most of these files are already compressed. The empty circles are the fast mode of paq8pxd and the previous cmz beta.

cmz 1.0 beta 5, measured 29 October 2026 · AMD Ryzen 5 3600

The files and what cmz makes of them

Result = size after compression as a share of the original, so smaller is better. Each file was packed, unpacked again and compared with the original byte by byte. Sizes are exact byte counts, so you can compare your own compressor directly.

#	File, source and SHA-256	Original, bytes	cmz result	Pack	Unpack
Text
01	English text: War and Peace (Tolstoy)war_and_peace.txt · Project Gutenberg #26002d5bb2ad5f422765e714617e21fa31bbaf8958aa79682c86fca6660fcc5d1b2b	3,359,610	21.5%720,928	1 min 24 s	16 s
02	Hungarian text: Az arany ember, vol. 1 (Jókai)az_arany_ember_1.txt · Project Gutenberg #56591d2118a7ba84c0433165e0b233221b08e2b18d589452f92b282ebedaf55bcabe8	555,710	28.2%156,934	14 s	2.8 s
03	EPUB e-book: Moby Dick (Melville)moby_dick.epub · Project Gutenberg #27011bc88e79ef7cf9d5325abe0071e6302265b708df56d3af07db124c81e84197f6	814,023	69.4%565,334	29 s	8.4 s
Documents
04	PDF form: Form 1040irs_f1040.pdf · US Internal Revenue Service3d31c226df0d189ced80e039d01cf0f8820c1019681a0f0ca6264de277b7e982	220,237	31.9%70,325	28 s	4.6 s
05	PDF publication: Publication 17irs_p17.pdf · US Internal Revenue Service2d2381d62c7c77ed4b64b83ac23c28ed85b7c8304eb580c09bd9fd660be0840b	3,088,632	36.7%1,132,942	2 min 46 s	1 min 23 s
06	PDF from an office suiteonlyoffice_sample.pdf · ONLYOFFICE document-templates52a4c59f9858efa9bbe6740bbacd1005f09c33db0aae11e39ff730f00a379bb6	1,329,569	74.8%994,660	34 s	4.2 s
07	Word documentonlyoffice_sample.docx · ONLYOFFICE document-templatesd678383aee1f2deb84808353fcf77e1ffd865495aa5079ff9bca90a24a9ca0f5	410,390	54.8%224,756	56 s	8.5 s
08	Excel workbookonlyoffice_sample.xlsx · ONLYOFFICE document-templates3d8a1ab5d84e5644995bd3384dd6c567ceeaacade34cf90cabd9d0bc430a5e41	76,825	19.6%15,060	24 s	2.1 s
09	PowerPoint presentationonlyoffice_sample.pptx · ONLYOFFICE document-templates524d1144395e5889f217ee9f84c7331433e2ac99735d933f82390ca9a49dcb5d	87,354	24.9%21,763	22 s	2.9 s
Images
10	JPEG photo: Apollo 11, AS11-40-5874nasa_as11-40-5874.jpg · NASA Image and Video Library9c93e64991da8c71352e8290f4faa57f1239bd87d74f956a8489a93d7201c8bb	2,236,993	72.1%1,612,870	5.9 s	2.5 s
11	PNG photo: Kodak test image 23kodim23.png · Kodak Lossless True Color Image Suite (r0k.us)e3111a2fd4da24af15d6459ef9eacfe54106b38e27b4a21821b75c3f5d2d5baf	557,596	82.9%462,003	26 s	6.4 s
12	TIFF image: Mandrill, 4.2.03sipi_4.2.03.tiff · USC-SIPI Image Database3f590b52279fb59b81906f1e928ae713a5357b1afc1a2017a103adb563fb4494	786,572	74.2%583,921	0.3 s	0.1 s
Programs and archives
13	Windows program: PuTTY 0.83, 64-bitputty_0.83_w64.exe · Simon Tatham (the.earth.li)16cbe40fb24ce2d422afddb5a90a5801ced32ef52c22c2fc77b25a90837f28ad	1,709,672	45.2%772,030	34 s	7.3 s
14	Source code, .tar.gz: curl 8.11.1curl-8.11.1.tar.gz · curl.sea889ac9dbba3644271bd9d1302b5c22a088893719b72be3487bc3d401e5c4e80	4,169,067	48.2%2,007,487	3 min 13 s	1 min 32 s
15	Zip archive: SQLite 3.47.2 amalgamationsqlite-amalgamation-3470200.zip · sqlite.orgaa73d8748095808471deaa8e6f34aa700e37f2f787f4425744f53fdd15a89c40	2,811,504	49.1%1,380,249	1 min 45 s	56 s
Data
16	SQLite database: Chinook samplechinook.sqlite · lerocha/chinook-database v1.4.5bdf635be69850bd3be09c9a2dbeef7ddfb80036bd3ef3381383cd03b61e4a61a	1,067,008	17.3%184,927	30 s	5.2 s
17	CSV table: airport codesairport-codes.csv · datasets/airport-codes (DataHub)e001037619f2cd5a2d3afe84ebbf83fc7b803edab6cc74f638342f1a078d9b43	8,822,393	16.4%1,450,020	1 min 36 s	51 s
18	JSON: countries of the worldcountries.json · mledoze/countries0d92fb23b43ae080bd439f5dc5d61a8845395fdd0af1479b433c59fe079bd6fd	1,398,004	5.3%73,475	29 s	6.5 s
19	Web server log: Apache, 2,000 linesapache_2k.log · Loghub (logpai)c7efa3eb686e3a96bd2f8f4457b2a7887e9cf2f3649327f1b4e87af841363ce8	171,239	2.2%3,709	3.2 s	3.1 s
Audio
20	WAV audio: Bach, March in D, pianobach_march_in_d.wav · Wikimedia Commons9d87a1d1183496452bd9b40e37a255ece7378821dabd8ae5a7b65dbe4f3dec2d	9,961,516	29.1%2,897,201	2.4 s	0.8 s
21	MP3 music: brass band, Symphony No. 3usafa_band_symphony3.mp3 · Wikimedia Commons (USAF Academy Band)c5f09c9ffe52762e6b556d79612ba1e07a3dacd57295e4146a09525807b755ba	7,801,849	79.8%6,224,812	35 s	12 s
22	MP3 speech: Pride and Prejudice, chapter 1librivox_pride_01_64kb.mp3 · LibriVox (archive.org)aa7339960c90bed4c57ffe144c96b7a8ef4de482d6505d19906bb50248fe6f0c	2,625,429	90.0%2,361,955	8.9 s	3.5 s
Science
23	FITS table: Kepler light curvekplr011904151-2009131105131_llc.fits · MAST, STScId14d76faf0bb4a245c1a728c5fbe3831c084570c0fed253b46d1b9b28837a337	77,760	22.9%17,817	3.0 s	1.5 s
24	FITS image: photographic plate scan, Oriondss_orion_10arcmin.fits · Digitized Sky Survey, STScI · the server writes the download date into the header, so a copy downloaded by hand has a different checksum; the script fixes it (see Notes)025b6059ea05e40051aa43c42191769cfc33ca5fc82387d4ce11a14af2f64c79	766,080	25.0%191,711	30 s	3.7 s
25	FITS image: the Sun at 171 Å, SDO/AIAAIA20240101_0000_0171.fits · JSOC, Stanford7bd72d03d357cd95a949b5d4294f7acfc251dd5436bac71d374690605f07333a	1,258,560	90.7%1,140,908	1.4 s	0.3 s
26	HDF5: LIGO strain data around GW150914H-H1_GWOSC_4KHZ_R1-1126259447-32.hdf5 · GWOSCc5ea87beced5094b56b4d694f3a36e6d59f74d3a46e546b980a6a199d1f252a9	1,040,781	82.7%861,242	23 s	5.2 s
27	Star catalogue, fixed-size binary recordssao · Silesia corpus (inside silesia.zip)c2d0ea2cc59d4c21b7fe43a71499342a00cbe530a1d5548770e91ecd6214adcc	7,251,944	53.3%3,863,259	1 min 40 s	31 s
	All 27 files	64,456,317	46.5%29,992,298	19 min 43 s	7 min 0 s

Times are wall-clock time on one PC (AMD Ryzen 5 3600, 6 cores, Windows 11), one file at a time; cmz uses several cores on large files. cmz ran with its default settings, the same ones the program uses when you pick Compress with cmz.

Download and check

Get the same files

The script makes a folder called cmz-test-files, downloads each file from the address in the list and prints OK or DIFFERENT for each one. Files that are already there and match are skipped, so you can run it again at any time.

Windows

Save get-test-files.ps1, then run it in PowerShell from the same folder:

powershell -ExecutionPolicy Bypass -File .\get-test-files.ps1

Tested on Windows 11. It is written to run on Windows 7 too, with the PowerShell that comes with it. A Windows 7 without updates may not be able to open the secure (TLS 1.2) connections most of these sites require; the script then says so, and Windows Management Framework 5.1 from Microsoft fixes it.

Linux, macOS, Git Bash

Save get-test-files.sh, then run:

bash get-test-files.sh

Needs curl, sha256sum or shasum, and unzip or bsdtar.

The files take 64.5 MB. The script downloads about 125.4 MB, because sao comes out of the 68 MB Silesia corpus zip, which it deletes afterwards.

Notes

Why some links look the way they do

Fixed versions. The ONLYOFFICE samples and the three files from GitHub repositories (CSV, JSON, log) are linked at a fixed commit, so they cannot change under you. The airport-codes CSV, for example, is updated every day.
The DSS image. The Digitized Sky Survey server makes the cutout on request and writes the download date into the FITS header. The image data itself is the same every time. The script writes back the date of our copy (2026-10-27), so the checksum matches. For a copy downloaded by hand, this PowerShell line does the same:
```
$p='dss_orion_10arcmin.fits'; $b=[IO.File]::ReadAllBytes((Resolve-Path $p).Path); [Text.Encoding]::ASCII.GetBytes('2026-10-27').CopyTo($b,491); [IO.File]::WriteAllBytes((Resolve-Path $p).Path,$b)
```
If a file shows DIFFERENT. A publisher may update a file, as Project Gutenberg sometimes does. The results on this page belong to the version with the checksum shown here.
Rights. Every file stays under its publisher's terms. We only link to it.
Your results. If you run another compressor on these files, we would like to hear about it, good news or bad: fmolnar.no1@gmail.com.