Readability Benchmark¶

Performance comparison between zerodep readability, readability-lxml, and Mozilla Readability.js.

Implementations¶

Test Fixtures¶

Benchmarks use real-world HTML fixtures from Mozilla Readability.js test suite, plus synthetic fixtures for controlled scaling:

Performance: zerodep vs readability-lxml¶

Performance: zerodep vs Mozilla Readability.js¶

All three implementations are now benchmarked via pytest-benchmark in the same CI run. Mozilla Readability.js is invoked through a Node.js subprocess.

Key Takeaways¶

• zerodep dominates on small and synthetic pages -- 2.5x faster on the small real-world fixture and 1.7-2.3x faster across all synthetic sizes vs readability-lxml. The optimized scoring and tree-walking algorithms shine on cleaner HTML.
• readability-lxml retains an edge on large complex pages -- lxml's C-based parser still provides an advantage on complex HTML like Wikipedia articles (1.2x slower).
• Massively faster than Mozilla Readability.js -- zerodep is 3.2-20x faster than the original JS reference across all fixtures. The JS implementation pays heavy overhead from jsdom's DOM construction.
• zerodep has richer metadata -- JSON-LD extraction, RTL support, and OpenGraph metadata that readability-lxml lacks
• Zero pip dependencies -- zerodep needs only the stdlib, while readability-lxml requires lxml and cssselect

Run It Yourself¶

# All benchmarks (zerodep vs readability-lxml vs Mozilla JS, requires Node.js)
pip install pytest pytest-benchmark readability-lxml
cd readability && npm install
pytest readability/test_readability_benchmark.py --benchmark-only -v

Latest CI Results¶

Updated automatically on each release via Benchmark CI.