zawk vs gawk

zawk and gawk run the same AWK language, and most programs give the same output in both. They are built for different jobs, though: gawk is the reference GNU AWK, available almost everywhere, while zawk is a fast AWK with a standard library for structured data and services. This page helps you pick one.

At a glance

	zawk	gawk
Language	AWK + gawk extensions + modern syntax	AWK + gawk extensions
Implementation	Rust, type inference, JIT with Cranelift	C, bytecode interpreter
Performance	usually 2-8x faster on large files, parallel with -p	single-threaded
CSV / TSV	RFC 4180 with -i csv, -i tsv, header names with -H	--csv since 5.3
Other input formats	JSON, JSON Lines, Parquet, Markdown tables, Excel	text only
Output formats	CSV, TSV, JSON Lines, Markdown, table, Parquet with -o	text only
Input sources	files, stdin, http(s)://, s3://, databases, NATS	files, stdin, /inet sockets
Standard library	~190 functions: JSON, date/time, crypto, HTTP, SQL, KV, LLM...	strings, math, time, bit operations
Extensions	built in, no loading needed	@load C extensions
Tooling	lint, fmt, inspect, functions, llm	--lint, --pretty-print, --profile, --debug
Arbitrary precision	no	-M with MPFR
Platforms	64-bit macOS, Linux, Windows	almost every Unix-like system, often preinstalled

Choose zawk when

• The data is structured: quoted CSV, JSON Lines, Parquet or Excel files, read with -i and addressed by column name instead of by position.

  zawk -i csv -H '{ total[$FI["city"]] += $FI["amount"] } END { for (c, t in total) print c, t }' orders.csv

• The files are large: type inference and machine code compilation make zawk faster than gawk on most workloads, and -p r / -p f spread the work over all cores. See the benchmarks.

• The script talks to other systems: HTTP APIs, S3, SQLite/MySQL/PostgreSQL, Redis/NATS KV, NATS/MQTT messages and LLMs, without shelling out to curl or a database client. See the standard library.

• You want a modern syntax: f-strings, ??, try ... else, match, local variables and for (k, v in arr). See New AWK Syntax.

• AI agents write or run the scripts: zawk inspect shows the schema of a file, zawk functions lists the library, zawk lint reports problems as JSON and zawk llm builds a prompt. See Tooling.

Choose gawk when

• The script must run anywhere without installing anything: gawk, or another POSIX AWK, is already on most Linux systems and in most container images.
• The script relies on gawk features zawk leaves out: arrays of arrays (a[i][j]), BEGINFILE/ENDFILE, FIELDWIDTHS/FPAT, RT, co-processes (|&) and /inet networking, @namespace, indirect function calls (@f()), @load extensions, or i18n with dcgettext().
• You need exact numbers beyond doubles: gawk's -M gives arbitrary precision arithmetic.
• The script assigns to $0, fields or NF to rebuild records, or edits ARGV to change the input files. These are read-only in zawk.
• You need gawk's debugger or profiler (--debug, --profile).
• Exact POSIX semantics matter, such as leftmost-longest regex alternation or treating "0" as a true string constant.

Using both

Most AWK one-liners and scripts run unchanged in zawk, so you don't have to pick once for all:

• Keep portable scripts in plain AWK, and run them with zawk where it is installed for speed.
• Run zawk lint -f script.awk when moving a gawk script to zawk: it reports most behavior differences with a dedicated code.
• Pipe them into each other: zawk reads Parquet, JSON or a database table and writes CSV or TSV that any AWK can process, and the other way round.

The full list of behavior differences is in Differences from gawk.