← Back to Horizon

Tuntau Horizon · benchmarks

How Horizon compares

Measured results for Horizon at every level against gzip, bzip2, xz, zstd, brotli, lz4, zip and 7-Zip. Every archive was unpacked again and checked byte for byte. The losses are in here too.

Also: try it with your own file or see results from other machines.

Test your own fileThese are my files. Drop an HTML, CSS, JS, JSON or image file here and compare Horizon with gzip, brotli and zstd on yours, live on my server.

In short

Smallest on every file

Normal against 7-Zip

Already compressed data

Where it loses

Size against speed

Up and to the right is better: smaller files, made faster. Each dot is one tool at one setting.

File by file

Size after packing, as a share of the original (shorter is better), for Horizon, 7-Zip and zstd.

How size scales

The same data cut to 1 MB, 10 MB, 100 MB and 1 GB. Most tools get relatively better with more data; the question is how fast.

All numbers

Click a column to sort. MB/s is megabytes of input per second; times are wall-clock, one run each.

All numbers

How this was measured

One machine, every tool at its strongest common setting, all CPU cores available (Horizon, zstd and xz use several threads). Each archive was unpacked again and compared with the original byte for byte. Times are single runs, so treat small differences as noise.

Other machines

My numbers come from one machine. Below is an estimate for your device, from the closest run people shared with horizon bench --upload. Every shared run, with filters, is under “All machines and filters”.

An upload holds the machine details and timings shown here, plus a random ID created on your computer so your repeat runs are merged; it is not linked to your hardware. No name, no file names, and the ID and IP address are never shown.

All machines and filters

Filter by hardware

Shared results per machine
#

Score: how quickly the data gets from one computer to another, counting packing, sending the packed file and unpacking, in megabytes of original data per second × 10. Higher is better. Tools are ranked only within their class, and each class is scored on the connection it is made for: Fast over 1 Gbit/s, Everyday over 100 Mbit/s, Smallest over 10 Mbit/s. So a size-first setting is never ranked against a speed-first one. Size is the packed size as a share of the original. Click a column to sort.