Languages Regex Benchmark

It's just a simple regex benchmark for different programming languages.

"All benchmarks are wrong, but some are useful" ~ Szilard (cited from benchm-ml)

We hope this benchmark can be helpful. If you take this as a guidance on choosing a language: Remember your choice should not only be about performance, but each language also has its engine and offers different features (like UTF support, backreferences, capturing groups ...).

Results

• 23 July 2026 (Latest)
• 01 April 2021

Summary of latest run

Language	Total(ms)
1. Rust	15.53
2. C++ SRELL	24.82
3. C# .Net	57.69
4. Nim Regex	59.50
5. PHP	70.66
6. Nim	70.84
7. Julia	108.77
8. Crystal	151.15
9. Javascript Node 26	158.21

Latest versions compared

Base image: ubuntu:24.04 (noble, amd64). Everything runs from a single Docker image whose entrypoint (benchmark/monolithic-from-sources/run-benchmarks.php, executed by PHP) builds and times each language.

• C: gcc 14.4.0 & PCRE2 10.47 (both built from source; C is compiled with gcc 14)
• Crystal: Crystal 1.21.0 (LLVM 22.1.0 bundled)
• C++: g++ 16.1.0 | Boost 1.91.0 (both built from source; C++ is compiled with g++ 16)
• C#: dotnet 10
• D: DMD 2.112.0 | LDC 1.42.0
• Dart: Dart 3.12.2
• Go: go 1.26.5
• Java: Eclipse Temurin OpenJDK 26.0.1+8
• Javascript: Node 24.18.0 (LTS) and Node 26.5.0 (Current) => two result rows
• Julia: Julia 1.12.6
• Kotlin: kotlinc-jvm 2.4.10
• Nim: Nim 2.2.10 + nim-regex 0.26.3
• Perl: perl 5.44.0 (built from source)
• PHP: PHP 8.5
• Python: Python 3.14 | PyPy 7.3.23 (PyPy2 & PyPy3)
• Ruby: ruby 4.0.6 (built from source)
• Rust: rustc 1.97.1 & regex 1.13.1

Note on exactness: Everything is pinned to an exact version except the three installed from apt feeds below, which are pinned to the minor version and track the latest patch (the feeds don't retain older patches, so an exact pin would break on rebuild):

• C# / .NET: dotnet-sdk-10.0 from the Microsoft prod feed → latest 10.0.x
• PHP: php8.5-cli from ppa:ondrej/php → latest 8.5.x
• Python 3: python3.14 from the deadsnakes PPA → latest 3.14.x

PHP 8.5 is reused as the benchmark orchestrator runtime.

Methodology & Parameters

Check each results/ file for their specific methodology. They might differ in the exact execution. However, we try to keep them comparable.

Input text

The input text is a concatenation of Learn X in Y minutes repository.

Feel free to open an issue for more/other text inputs.

Regex patterns

• Email: [\w\.+-]+@[\w\.-]+\.[\w\.-]+
• URI: [\w]+://[^/\s?#]+[^\s?#]+(?:\?[^\s#]*)?(?:#[^\s]*)?
• IPv4: (?:(?:25[0-5]|2[0-4][0-9]|[01]?[0-9][0-9])\.){3}(?:25[0-5]|2[0-4][0-9]|[01]?[0-9][0-9])

The above regex patterns aren't the best nor the optimal. The focus is the benchmark, not the matching.

The patterns are applied to the whole file.

Measure

Measuring is done inside the programs to avoid including startup, reading and writing times on results.

Elapsed time includes pattern compilation, find and count occurrences.

Usually, an average out of 10 runs is measured.

How to run

We serve two ways of running the benchmark. One is a fully isolated docker container running all of them, being able to offer a fully equalised OS base. The other is hosting one container per language, reusing existing images for most of the languages (accompanied by some custom images), which is much faster but not as equalised on the OS-level, tho we try to keep it as similar as possible. They are independent and only hand-synced.

Option A (Fully isolated custom ubuntu image)

docker build -t regex-benchmark benchmark/monolithic-from-sources
docker run --rm -v $(pwd):/var/regex regex-benchmark

Note: the image builds GCC 14, GCC 16, PCRE2, Boost, Perl and Ruby from source, so the initial docker build can take a long time (potentially over an hour on a typical machine) and needs several GB of disk. Subsequent builds reuse the Docker layer cache.

Option B (One image per language)

docker compose -f benchmark/from-images/docker-compose.yml up --build

See its README for the per-language tags and the fairness caveats when comparing its table to the single-image one.

Contributing

All contributions are welcome, from tiny optimizations to new implementations.

There are only a few requirements:

• Follow the style of the current implementations
• Use the default settings for the regex engine
• Update Dockerfile if it's necessary

Kudos

• mariomka and all contributors for their work on the benchmark
• Heng Li's for his work on Benchmark of Regex Libraries
• A "challenge" on Madrid Devs group, who inspired mariomka for the original benchmark
• Programming subreddit, helped mariomka to improve the original benchmark

License

MIT © Mario Juárez.