pdfplumber vs PyMuPDF

Plain text

import pdfplumber
import pymupdf  # imported as "fitz" in older code

path = "samples/invoice.pdf"

with pdfplumber.open(path) as pdf:
    print("pdfplumber:", pdf.pages[0].extract_text().splitlines()[:3])

with pymupdf.open(path) as doc:
    print("PyMuPDF:   ", doc[0].get_text().splitlines()[:3])

pdfplumber: ['Hollow Pine Supply Co. INVOICE', '418 Quarry Road, Unit 6', 'Invoice no. INV-2026-0417']
PyMuPDF:    ['Hollow Pine Supply Co.', '418 Quarry Road, Unit 6', 'Millbrook, OR 97000']

Both return the same words. The order differs: pdfplumber groups by line across the page (so INVOICE, far right, joins the company name), while PyMuPDF follows the blocks in the order the PDF draws them. On multi-column pages that difference decides who gets reading order right with defaults: PyMuPDF did, in our test. pdfplumber needs a crop per column (how).

Speed is PyMuPDF’s clearest win: about 28× faster on the 15-page document. PyMuPDF wraps the MuPDF C library; pdfplumber is Python on top of the pure-Python pdfminer.six.

Words and coordinates

import pdfplumber
import pymupdf

path = "samples/invoice.pdf"

with pdfplumber.open(path) as pdf:
    w = pdf.pages[0].extract_words()[0]
    print("pdfplumber:", w["text"], round(w["x0"], 1), round(w["top"], 1), round(w["x1"], 1), round(w["bottom"], 1))

with pymupdf.open(path) as doc:
    x0, y0, x1, y1, text, *_ = doc[0].get_text("words")[0]
    print("PyMuPDF:   ", text, round(x0, 1), round(y0, 1), round(x1, 1), round(y1, 1))

pdfplumber: Hollow 54.0 56.1 119.6 76.1
PyMuPDF:    Hollow 54.0 50.6 119.6 78.1

Both use a top-left origin and points, so x0, x1 match. The vertical edges differ: pdfplumber’s box is exactly as tall as the font size (20 pt here, pdfminer.six’s convention), while PyMuPDF’s spans the font’s ascender to descender (27.5 pt). Neither is wrong; just never mix coordinates from the two libraries in one calculation.

Regions

import pdfplumber
import pymupdf

path, box = "samples/invoice.pdf", (50, 125, 300, 225)   # x0, top, x1, bottom

with pdfplumber.open(path) as pdf:
    print("pdfplumber:", pdf.pages[0].crop(box).extract_text().splitlines())

with pymupdf.open(path) as doc:
    print("PyMuPDF:   ", doc[0].get_text(clip=pymupdf.Rect(*box)).splitlines())

pdfplumber: ['BILL TO', 'Larkspur Robotics GmbH', 'Attn: Accounts Payable', 'Feldstraße 12', '20357 Hamburg, Germany']
PyMuPDF:    ['BILL TO', 'Larkspur Robotics GmbH', 'Attn: Accounts Payable', 'Feldstraße 12', '20357 Hamburg, Germany']

Identical results. PyMuPDF’s clip= works on text; pdfplumber’s crop() returns a whole page object, so tables, lines, chars and filter() all work inside the region too (crop and coordinates).

Tables

import pdfplumber
import pymupdf

path = "samples/invoice.pdf"

with pdfplumber.open(path) as pdf:
    print("pdfplumber:", pdf.pages[0].extract_table()[1])

with pymupdf.open(path) as doc:
    table = doc[0].find_tables().tables[0]
    print("PyMuPDF:   ", table.extract()[1])

pdfplumber: ['HP-1042', 'Stainless hex bolts M8 x 40 (box of 100)', '12', '18.40', '220.80']
PyMuPDF:    ['HP1042\n-', 'StainlesshexboltsM8x40(boxof100)', '12', '1840\n.', '22080\n.']

Both found the table and its cells. With PyMuPDF 2.1, cell text lost its spaces and punctuation moved to a separate line, on every table we tried, including the FBI table from pdfplumber’s test suite. PyMuPDF’s table finder has a strategy="text" option for borderless tables; in our test it did no better. pdfplumber’s table settings and visual debugger are more mature, and with the settings from our guides it reproduced all three test tables exactly (results).

Lines, rectangles and curves

import pdfplumber
import pymupdf

path = "samples/invoice.pdf"

with pdfplumber.open(path) as pdf:
    page = pdf.pages[0]
    print("pdfplumber: lines", len(page.lines), "| rects", len(page.rects), "| first line", page.lines[0]["x0"], page.lines[0]["top"])

with pymupdf.open(path) as doc:
    drawings = doc[0].get_drawings()
    items = [item[0] for d in drawings for item in d["items"]]
    print("PyMuPDF:    drawings", len(drawings), "| path items", {k: items.count(k) for k in set(items)})

pdfplumber: lines 15 | rects 1 | first line 54.0 250.0
PyMuPDF:    drawings 16 | path items {'re': 1, 'l': 15}

pdfplumber exposes vector graphics as ready-to-use objects with x0, top, x1, bottom, colours and widths, which is what its table finder and your own geometry code consume. PyMuPDF’s get_drawings() returns path descriptions ('l' for line, 're' for rectangle, 'c' for curve) that you interpret yourself.

What only one of them does

fitz and pymupdf

Old tutorials use import fitz. That is the same library: PyMuPDF’s original module name, kept for compatibility.

import pymupdf

try:
    import fitz
    print("import fitz works; same module:", fitz.Document is pymupdf.Document)
except ImportError as e:
    print("import fitz failed:", e)
print("PyMuPDF", pymupdf.VersionBind)

import fitz works; same module: True
PyMuPDF 2.1

Use import pymupdf in new code. There is an unrelated PyPI package called fitz; if pip install fitz was run by mistake, uninstall it and install pymupdf.

The licence question

pdfplumber is MIT: use it in any product, open or closed, without obligations beyond keeping the copyright notice. PyMuPDF is AGPL-3.0: if you distribute software that uses it, or let users interact with it over a network (a SaaS document pipeline counts), you must offer your complete source code under the AGPL, or buy a commercial licence from Artifex. Version 2.1 also installs pymupdf4llm and onnxruntime, which is why it measured 282.9 MB with dependencies.

Using both

The libraries combine well: PyMuPDF for fast text and rendering across a whole collection, pdfplumber for the pages where you need tables or precise geometry. Open the same file with each; they do not interfere. Keep each library’s coordinates within its own calculations.