Comprehensive Benchmarking of Python Text Extraction Libraries in 2025: Results and Insights

In the rapidly evolving landscape of Python-based document processing, choosing the right text extraction library is crucial for efficiency, reliability, and scalability. Recent benchmarking efforts have shed light on the performance of four prominent Python libraries across a wide spectrum of real-world documents. This detailed analysis aims to help developers, data scientists, and enterprise teams make informed decisions based on empirical data.

Exploring the Leading Python Text Extraction Tools

The benchmarking study evaluates the following libraries:

• Kreuzberg: An open-source, lightweight library designed for fast and reliable text extraction.
• Docling: A robust solution leveraging machine learning to understand complex document layouts.
• MarkItDown: A straightforward tool optimized for converting simple PDFs and Office documents to Markdown.
• Unstructured: An enterprise-grade framework tailored for processing diverse and complex document types.

Test Methodology and Scope

The evaluation encompasses 94 documents varying significantly in size—from tiny files under 100 KB to massive academic papers exceeding 50 MB. These documents span multiple formats, including PDFs, Word documents, HTML, images, and spreadsheets, and cover various languages such as English, Hebrew, German, Chinese, Japanese, and Korean. To ensure fairness, all tests were conducted on CPU-only systems, with no GPU acceleration involved.

Key performance metrics assessed included processing speed, memory consumption, success rates, and installation footprint. Multiple runs per document provided statistically significant insights, while open-source scripts and datasets guarantee transparency and reproducibility.

What the Results Tell Us

Speed and Efficiency

• Kreuzberg emerged as the fastest, processing over 35 files per second, demonstrating remarkable efficiency across all document types and sizes.
• Unstructured also performed reliably, achieving substantial processing throughput suitable for enterprise environments.
• MarkItDown handled simple documents adeptly but struggled with complex or large files.
• Docling, despite its advanced ML capabilities, often required over an hour per document and frequently timed out on medium-sized files, limiting its practicality for large-scale workflows.

Installation and Resource Usage

• Kreuzberg’s minimal footprint (around 71 MB with 20 dependencies) makes it highly suitable for deployment in resource-constrained environments like cloud functions.
• Unstructured requires about 146 MB and 54 dependencies, offering a good balance between capability and resource demands.
• MarkItDown’s installation size is 251 MB, mainly due to additional components like ONNX.
• Docling’s hefty 1 GB size, combined with