Firecrawl rewrote its PDF parsing engine using Rust, increasing speed by up to 5.7 times.

According to ME News, on April 15th (UTC+8), 1M AI News reported that Firecrawl, a web data extraction tool, released Fire-PDF. This PDF parsing engine, rewritten in Rust, converts PDFs to structured Markdown 3.5 to 5.7 times faster than its predecessor, with an average processing time of less than 400 milliseconds per page. The key to this speedup lies in reducing unnecessary GPU usage. Firecrawl also open-sourced the Rust library pdf-inspector, which can classify each PDF page in milliseconds: plain text pages are extracted directly natively, bypassing the GPU; only scanned documents or image-intensive pages are processed using a neural network layout model and a GLM-OCR visual language model. For example, in a 150-page text and 60-page scanned financial report, most pages require no GPU input. Regarding accuracy, Fire-PDF sets different parameters for different content types: tables receive higher token limits and a maximum generation time of 25 seconds; formulas are preserved in LaTeX; and multi-column layouts use neural networks to predict the reading order. Fire-PDF is automatically enabled for all Firecrawl users; no configuration is required. (Source: ME)