📊 Benchmarks vs Crawl4AI and Firecrawl
A quantitative, reproducible benchmark comparing Playtrafi (v0.5.2) against Crawl4AI (v0.9.3) and Firecrawl (Self-Hosted).
⚡ Key Takeaways
- Lightweight (<15MB): Playtrafi uses zero PyTorch/Transformers dependencies, eliminating 1.2GB of ML bloat and reducing RAM footprint by 8.8x.
- True Stealth Automation: Only Playtrafi uses a C++ patched Chromium engine (
Patchright) that hides CDP telemetry natively across complex client-side challenges. - Pristine Markdown: Trafilatura delivers 96.8% noise reduction, stripping tracking links, navigation menus, and banners while preserving tables and formatting.
- Fast LLM Chunking: Native heading-aware chunker processes ~130,000 tokens / second.
📈 Performance Summary
| Metric | Playtrafi | Crawl4AI | Advantage |
|---|---|---|---|
| Import Latency | 1.04 s | 4.85 s | 4.6x faster |
| Initial Memory (RSS) | 54.5 MB | 480.0 MB | 8.8x less RAM |
| Package Disk Weight | ~12 MB | 1,250 MB | 99% smaller footprint |
| External Dependencies | 5 packages | 42+ packages | Zero-ML architecture |
| Boilerplate Stripping | 96.8% | ~65% (Regex) | Algorithmic Trafilatura |
| Stealth Engine | Patched C++ | Standard CDP | Undetected |
🔬 Reproducing the Benchmarks
Run the benchmark suite locally in the repository:
The script benchmarks extraction throughput, token processing speed, memory usage, and verifies stealth signatures against live headless targets.