What Is a .ZST File?

Zstandard Compressed File (.zst)

Overview

Zstandard (zstd) is a lossless compression algorithm and file format developed by Yann Collet at Facebook and released in 2016. A .zst file is a single stream of Zstandard-compressed data, in the same way a .gz file is a gzip stream: it compresses one file, and archives of many files are .tar.zst.

Zstandard's appeal is the trade-off it offers. At its default level it compresses about as well as zlib while decompressing three to five times faster; at high levels it approaches xz's ratio at a fraction of the time. That combination made it the default in a remarkable number of places in under a decade: Linux kernel modules and initramfs, Arch and Fedora packages, Btrfs and ZFS transparent compression, PyTorch and Parquet, and the --zstd option in tar, rsync and git.

Technical Specifications

• MIME Type: application/zstd
• File Extensions: .zst; .tzst or .tar.zst for tarballs
• Category: Archive
• Specification: RFC 8878
• Algorithm: LZ77-style matching with finite state entropy (tANS) and Huffman coding
• Levels: 1 to 19 normally, 20 to 22 with --ultra; negative levels for extreme speed
• Released: August 2016
• Licence: BSD and GPLv2, dual
• Magic bytes: 28 B5 2F FD

Identification

00000000  28 b5 2f fd 64 00 00 ...                          |(./.d...|

Every Zstandard frame begins with 28 B5 2F FD. A file may contain several frames back to back. Skippable frames, used for metadata and seek tables, begin 50 2A 4D 18 through 5F 2A 4D 18 and decoders ignore them.

File Structure

Zstandard frame
├── Magic number            28 B5 2F FD
├── Frame header            2 to 14 bytes: content size, window size, dictionary id, checksum flag
├── Data blocks             each with a 3-byte header: last-block flag, type (raw / RLE / compressed), size
└── Content checksum        optional 4-byte XXH64 fragment

Blocks are at most 128 KB. The window size in the header tells the decoder how much memory it needs, which is why a file compressed with --long at a huge window may refuse to decompress on a small device unless the same flag is given.

Dictionaries

Zstandard can be trained on a sample of similar small files to produce a dictionary that dramatically improves compression of that kind of data (log lines, JSON records, protocol messages). A .zst file compressed with a dictionary records the dictionary id and cannot be decompressed without it, which occasionally surprises people handling files from a database or message queue.

History and Development

Collet, who had earlier written LZ4, designed Zstandard to fill the gap between LZ4's speed and xz's ratio. Facebook open-sourced it in 2016 and uses it across its infrastructure. The IETF published the format as RFC 8478 in 2018 and the revised RFC 8878 in 2021.

Adoption milestones: Linux 5.9 gained zstd kernel image compression (2020); Fedora switched RPM payloads to it (2019) and Arch its packages (2020); Ubuntu and Debian's apt and dpkg added support; Chrome and Firefox added Content-Encoding: zstd for HTTP (2024); Microsoft added it to Windows 11's built-in archive support (2023).

Common Use Cases

• Package managers: Arch, Fedora, openSUSE and others store packages as .tar.zst or zstd-compressed payloads
• Linux kernel and initramfs compression
• Filesystems: Btrfs and ZFS use zstd for transparent compression
• Backups and snapshots: tar --zstd, Borg, restic and ZFS send streams
• Data engineering: Parquet, ORC, Arrow and Kafka all offer zstd as a codec
• HTTP compression for responses to modern browsers
• Game assets and databases (RocksDB, MongoDB WiredTiger)

How to Open a ZST File

Linux and macOS

The zstd package provides the tool; it is in every distribution's repositories and in Homebrew:

# Decompress to the same name without .zst
zstd -d file.zst
unzstd file.zst            # same thing

# To standard output, for piping
zstdcat file.zst | less

# A tarball, in one step
tar --zstd -xf archive.tar.zst
tar -xf archive.tar.zst    # modern tar detects it from the magic bytes

# Inspect without extracting
zstd -l archive.tar.zst

Windows

• Windows 11 (23H2 and later): File Explorer opens .zst and .tar.zst natively; right-click, Extract All.
• 7-Zip 21.07 and later support zstd through the 7-Zip-Zstandard fork or built-in in recent versions; PeaZip and NanaZip support it fully.
• The official zstd.exe from the project's GitHub releases works in PowerShell exactly as above.

Graphical, any platform

• 7-Zip, PeaZip (Windows and Linux), Keka and The Unarchiver (macOS), File Roller and Ark (Linux).

In code

import zstandard as zstd          # pip install zstandard

with open("data.zst", "rb") as f, open("data", "wb") as out:
    zstd.ZstdDecompressor().copy_stream(f, out)

Node (@mongodb-js/zstd, fzstd), Go (klauspost/compress), Rust (zstd crate), Java (zstd-jni) and most other languages have bindings; Python 3.14 added compression.zstd to the standard library.

How to Create and Convert ZST Files

# Compress a file (level 3 default); keeps the original
zstd file

# Higher compression, remove the original
zstd -19 --rm file

# Fastest useful setting, multi-threaded
zstd -1 -T0 largefile

# Long-range matching for very large, repetitive inputs
zstd --long=27 -T0 backup.img

# Tarball
tar --zstd -cf archive.tar.zst folder/

# Convert from gzip or xz without an intermediate file
gzip -dc old.tar.gz | zstd -o new.tar.zst
xz -dc old.tar.xz | zstd -19 -o new.tar.zst

Choosing a level: 1 to 3 for speed-critical paths, 3 (default) for general use, 9 to 12 for a good balance when compressing once and reading often, 19 for distribution where every byte counts. Decompression speed barely changes with level, which is the property that makes high levels practical.

Advantages

• Very fast decompression at every level.
• Wide range: one tool spans LZ4-like speed to xz-like ratio.
• Multi-threaded compression built in (-T0).
• Dictionary mode for small records.
• Standardised (RFC 8878) and ubiquitous in modern systems.

Limitations

• Single stream: .zst holds one file; use tar for directories.
• Not universal on older systems: Windows 10 and older macOS need a tool installed.
• Dictionary-compressed files need the dictionary.
• Slightly behind xz at maximum ratio, for workloads where time is free.

Related Formats

• GZIP: the format zstd most often replaces.
• XZ: the highest-ratio alternative, much slower.
• LZ4: the same author's speed-first format.
• TAR: the archive format wrapped inside .tar.zst.
• PARQUET: a data format that uses zstd internally.