PDF Parser Benchmark
Five PDF parsers, one harness, the same documents.
Why this exists
Every RAG system starts with a parser, and almost nobody measures the one they picked. The default choice is whichever library the first tutorial used, and the quality ceiling of everything downstream is set right there.
So: five parsers, the same documents, one harness. PyPDF, PDFPlumber, PDFMiner, LlamaParse and AWS Textract, compared on extraction fidelity across layouts that actually break things — multi-column text, tables, and scanned pages.
The comparison is the artifact
The point is not that one parser wins. It is that the right parser depends on the document, and that this is a measurable question rather than a matter of taste. The harness is reusable: point it at your own corpus and it will tell you which one to use.