⚡ Milliseconds.dev — Benchmarks

.NET is seriously fast. Here's the proof.

We rewrite the world's most popular code in .NET and measure exactly how much faster it runs — against Python, Node.js, Ruby, Go, or whatever the original is written in. Real hardware. Real numbers. Sub-millisecond proof.

👉 milliseconds.dev

---

📦 What's in this repo

13 standalone C# / .NET 10 benchmark projects. Each folder contains:

XX-name/
  dotnet/      ← the C# implementation (.csproj + source)
  python/      ← the competing implementation
  data/        ← input datasets used in the benchmark
  RESULTS.md   ← timing results on reference hardware
  article.md   ← full write-up published on milliseconds.dev

---

🏆 Benchmarks

#	Benchmark	.NET impl	vs	Peak speedup
01	Softmax · Llama-3 vocab scale	TensorPrimitives	NumPy	1.5×
02	PageRank · 28.5M edges	CSR sparse + SIMD	NetworkX	47×
03	PDF text extraction · 200 papers	PdfPig	pypdf	6×
04	Tokenizer · 100 MB corpus	Compiled Regex	NLTK	8×
05	Full-text search · 1k BM25 queries	Lucene.NET	Whoosh	22×
06	Edit distance · 100k pairs	ArrayPool<int>	textdistance	71×
07	Markdown rendering · 10k docs	Markdig	mistune	11×
08	QR code generation · 50k codes	QRCoder	qrcode	14×
09	Date parsing · 1M timestamps	DateTimeOffset	dateutil	19×
10	Diff algorithm · 50k pairs	Custom DP LCS	difflib	21×
11	Interval tree · 1M intervals	Augmented BST	intervaltree	31×
12	Image pipeline · 10k images	ImageSharp	scikit-image	7.7×
13	HTML parsing · 50k documents	AngleSharp	BeautifulSoup4	17.4×

---

🚀 Running a benchmark

Each project is a self-contained .NET console app:

cd 06-levenshtein/dotnet
dotnet run -c Release

Requirements: .NET 10 SDK — download

To run the competing implementation:

cd 06-levenshtein/python
pip install -r requirements.txt
python benchmark.py

---

📊 Methodology

• All timings are wall-clock time, median of ≥ 3 cold runs
• .NET is compiled in Release mode (-c Release)
• Both implementations use the same input data from data/
• Benchmarks run on the same physical machine — specs in each RESULTS.md
• No micro-optimisation tricks that wouldn't appear in real production code

---

🤝 Contributing

Interested in adding a benchmark — any language, not just Python?

1. Pick a matchup where .NET has a clear equivalent
2. Write both implementations (idiomatic code, no hand-tuned tricks)
3. Benchmark on real hardware, report median of ≥ 3 runs
4. Open a PR with your dotnet/, competitor folder, and RESULTS.md

Or submit directly at milliseconds.dev/contribute — login with your email, attach your files, and we'll reach out.

---

Built by Mo Mand · milliseconds.dev