Firecrawl rewrites the PDF parsing engine in Rust, boosting speed up to 5.7 times faster than before

ME News Report, April 15 (UTC+8), according to 1M AI News monitoring, the web data extraction tool Firecrawl released Fire-PDF, a PDF parsing engine rewritten in Rust, which speeds up converting PDFs to structured Markdown by 3.5 to 5.7 times compared to the previous generation, with an average processing time per page below 400 milliseconds.
The speed improvement mainly comes from reducing unnecessary GPU calls. Firecrawl also open-sourced the Rust library pdf-inspector, which can classify each PDF page in milliseconds: pure text pages are directly extracted natively, skipping GPU; only pages with scans or dense images are sent to neural network layout models and GLM-OCR visual language models for processing.
For example, a financial report with 150 pages of text and 60 pages of scans mostly do not require GPU.
In terms of accuracy, Fire-PDF sets parameters for different content types: tables are given higher token limits and a maximum generation time of 25 seconds, formulas are preserved in LaTeX, and multi-column layouts are predicted for reading order using neural networks.
Fire-PDF is automatically enabled for all Firecrawl users without any configuration.
(Source: BlockBeats)