This FREE Tool Turns ANY PDF into Perfect Markdown (MinerU Live Test)

Every RAG pipeline dies at the same spot: PDF parsing. Tables turn to mush, formulas vanish, two-column layouts get scrambled — and then everyone wonders why retrieval quality is bad. In this video I install MinerU — the open-source document parsing engine with 73,000+ GitHub stars — on my own Windows laptop and put it through a real test: parsing Attention Is All You Need into clean, LLM-ready Markdown.

What MinerU is

MinerU (from OpenDataLab) is a document parsing pipeline, not just an OCR tool. It runs four stages:

Layout analysis — detect columns, headers, figures, reading order
OCR — extract text where needed
Formula recognition — math comes out as LaTeX, not garbled Unicode
Table extraction — real tables with structure intact

Multiple backends, and it fits consumer hardware: everything here ran on an RTX 4060 laptop, 8 GB VRAM.

Install

One line with uv, shown live in the video. Real terminal output, real parse times — no faked numbers.

The test: a genuinely hard paper

Attention Is All You Need is the perfect stress test: two-column layout, dense equations, multi-part tables. MinerU's output:

Formulas → clean LaTeX that renders
Tables → structured Markdown/HTML tables, rows and columns preserved
Layout → correct reading order across the two columns

There's also a built-in Gradio Web UI if you'd rather drag-and-drop than script it.

Pros and cons

Pros

Free, open source, massive community
LaTeX formula extraction is best-in-class for local tools
Runs on an 8 GB laptop GPU
Batch-friendly: perfect for building RAG corpora

Cons

Model downloads are chunky on first run
Complex scanned documents still trip it occasionally
GPU strongly recommended — CPU parsing is slow

Verdict

If your documents feed an LLM, garbage parsing caps everything downstream. MinerU is the best free answer I've tested: one install, and PDFs — even equation-heavy academic ones — come out as Markdown your pipeline can actually use. Full live parse in the video. Docs: opendatalab.github.io/MinerU.