Best Python libraries for reading and filling PDF files (worth paying for?)

Requested by a visitor · October 26, 2026

We asked 10 models: Which Python PDF libraries are worth using? Each named its own top 10.

Asked 4 ways: List request · Buyer asking for advice · Search query · Worth paying for?. Share of voice pools every wording.

Agreement between models: 62%
Same model, run twice: 73%
Different picks: 20

When agreement between models is well below a model's agreement with itself, the split reflects real differences in taste rather than randomness.

The consensus

We asked 10 models: Which Python PDF libraries are worth using? Each named its own top 10.

Each cell is the position that model gave the pick; blank means it left the pick off. Click a row for each model's reasoning.

#	Pick	Listed by	Claude Haiku 5.5	GPT-6 Luna	Gemini 4 Argon	Qwen 3.8 Flash	DeepSeek V4.1 Flash	GLM-5.3	Kimi K3	MiniMax M3	Mistral Large 4	Nemotron 3.5 Super VL
1	PyMuPDF	10/10	3	1	1	3	2	1	1	4	1	5
GPT-6 Luna (#1) Fast, feature-rich PDF rendering, text extraction, annotations, and document manipulation. Gemini 4 Argon (#1) Blazingly fast C-powered library for reading, rendering, and extracting text or images. GLM-5.3 (#1) Fastest rendering and extraction, excellent text/image handling. Kimi K3 (#1) Fastest and most capable for text extraction, rendering, and annotations. Mistral Large 4 (#1) Fastest and most versatile library for reading, editing, and rendering PDFs. DeepSeek V4.1 Flash (#2) Fastest, most capable rendering and text/image extraction; AGPL caveat. Claude Haiku 5.5 (#3) Fast, feature-rich rendering, extraction, and editing; strong all-around choice. Qwen 3.8 Flash (#3) Fastest full-featured engine; rendering, text, images, annotations. MiniMax M3 (#4) Fast and feature-rich for rendering, extraction, and manipulation Nemotron 3.5 Super VL (#5) Fast rendering, text extraction, and manipulation.
2	pypdf	10/10	2	2	3	1	1	2	2	3	5	6
Qwen 3.8 Flash (#1) Default pure-Python choice: merge, split, extract text reliably. DeepSeek V4.1 Flash (#1) Best default for splitting, merging, and text extraction; pure Python, actively maintained. Claude Haiku 5.5 (#2) Pure-Python, actively maintained, ideal for splitting, merging, and basic manipulation. GPT-6 Luna (#2) Best default for pure-Python PDF reading, writing, merging, splitting, and metadata tasks. GLM-5.3 (#2) Pure-Python standard for merging, splitting, encryption; actively maintained. Kimi K3 (#2) Actively maintained standard for merging, splitting, and manipulating PDFs. Gemini 4 Argon (#3) The reliable pure-Python standard for merging, splitting, and basic PDF manipulation. MiniMax M3 (#3) Modern actively maintained fork of PyPDF2 with better Python 3 support Mistral Large 4 (#5) Pure-Python, dependency-free successor to PyPDF2 for merging, splitting, and basic manipulation. Nemotron 3.5 Super VL (#6) Modern PyPDF2 fork with improved maintenance and features.
3	pdfplumber	10/10	4	3	2	2	3	3	3	2	3	4
Gemini 4 Argon (#2) Unbeatable visual debugging and accuracy for extracting complex tables and layout data. Qwen 3.8 Flash (#2) Best-in-class table and layout extraction from complex PDFs. MiniMax M3 (#2) Excellent for extracting text and tables from PDFs reliably GPT-6 Luna (#3) Excellent layout-aware text and table extraction with inspectable character-level details. DeepSeek V4.1 Flash (#3) Unmatched for table extraction and character-level layout debugging. GLM-5.3 (#3) Great for table extraction and detailed layout inspection. Kimi K3 (#3) Best-in-class table and text extraction with precise layout control. Mistral Large 4 (#3) Best-in-class table and text extraction with excellent accuracy for structured data. Claude Haiku 5.5 (#4) Excellent for extracting text, tables, and coordinates from text-based PDFs. Nemotron 3.5 Super VL (#4) Superior table and precise text extraction.
4	ReportLab	10/10	1	4	4	4	4	4	4	1	2	3
Claude Haiku 5.5 (#1) Mature, powerful library for generating complex PDFs programmatically. MiniMax M3 (#1) Most powerful for programmatic PDF generation with precise layout control Mistral Large 4 (#2) Industry standard for programmatic PDF generation with unmatched control and reliability. Nemotron 3.5 Super VL (#3) Powerful programmatic PDF generation from scratch. GPT-6 Luna (#4) Mature, dependable choice for generating PDFs programmatically. Gemini 4 Argon (#4) The industry heavyweight for programmatically generating complex PDFs from scratch. Qwen 3.8 Flash (#4) Most mature library for generating PDFs programmatically. DeepSeek V4.1 Flash (#4) The standard for programmatic PDF generation; precise layout control. GLM-5.3 (#4) Best choice for generating complex PDFs from scratch. Kimi K3 (#4) The go-to for programmatically generating complex PDF documents.
5	PDFMiner.six	8/10	5		9	5	6	5	5		6	2
Nemotron 3.5 Super VL (#2) Excellent text extraction and content analysis. Claude Haiku 5.5 (#5) Deep layout analysis and precise text extraction, though slower and lower-level. Qwen 3.8 Flash (#5) Fine-grained layout parsing; foundation for other tools. GLM-5.3 (#5) Reliable, precise text extraction with layout analysis. Kimi K3 (#5) Deep low-level text extraction when layout analysis matters most. DeepSeek V4.1 Flash (#6) Deep layout analysis and text extraction; good foundation, slower than alternatives. Mistral Large 4 (#6) Deep, detailed text extraction with layout analysis for complex documents. Gemini 4 Argon (#9) Deep layout analysis engine ideal for custom text extraction pipelines.
6	WeasyPrint	8/10	7	6	5	9	8	8		5	7
Gemini 4 Argon (#5) Best solution for converting HTML and CSS templates into beautifully styled PDFs. MiniMax M3 (#5) Best HTML/CSS to PDF conversion with solid standards compliance GPT-6 Luna (#6) Polished HTML and CSS to PDF when print-oriented web layout is the goal. Claude Haiku 5.5 (#7) Converts HTML and CSS to PDF well, great for styled report generation. Mistral Large 4 (#7) Turns HTML/CSS into beautiful PDFs, ideal for reports and styled documents. DeepSeek V4.1 Flash (#8) Best HTML/CSS-to-PDF rendering; strong for reports from templates. GLM-5.3 (#8) Excellent for HTML/CSS-to-PDF document generation.
7	pikepdf	8/10	8	7	6	7	5	6	8		9
DeepSeek V4.1 Flash (#5) Robust low-level edit and repair layer built on qpdf; great for metadata and structure. Gemini 4 Argon (#6) QPDF-backed powerhouse for low-level PDF repair, inspection, and lossless manipulation. GLM-5.3 (#6) Powerful qpdf-based editing, repair, and metadata manipulation. GPT-6 Luna (#7) Strong low-level editing and repair, backed by qpdf for robust PDF structure handling. Qwen 3.8 Flash (#7) qpdf-backed repair, encryption, metadata editing. Claude Haiku 5.5 (#8) Solid QPDF-based library for low-level structure, repair, and encryption work.
8	PyPDF2	5/10	6	9						6	4	1
Nemotron 3.5 Super VL (#1) Basic read/write, merge, split; easy to start with. Mistral Large 4 (#4) Legacy but still widely used for basic PDF operations in existing codebases. MiniMax M3 (#6) Pure Python library good for basic read, merge, split operations
9	fpdf2	7/10	10		8	6		7	7	9		9
Qwen 3.8 Flash (#6) Lightweight, dependency-free PDF creation, great for simple docs. GLM-5.3 (#7) Lightweight, easy PDF creation with good Unicode support. Kimi K3 (#7) Simple, dependency-free PDF generation for basic documents. Gemini 4 Argon (#8) Lightweight, easy-to-learn pure Python library for simple document generation. MiniMax M3 (#9) Lightweight library for simple PDF generation from scratch Nemotron 3.5 Super VL (#9) Simple, lightweight PDF creation. Claude Haiku 5.5 (#10) Lightweight, simple API for creating PDFs from scratch without heavy dependencies.
10	borb	6/10		5	10	10	7	9	6
GPT-6 Luna (#5) Modern toolkit for PDF generation and manipulation with an accessible Python API. Kimi K3 (#6) Modern unified API for both reading and creating PDFs. DeepSeek V4.1 Flash (#7) Solid pure-Python library for reading, writing, and styling PDFs with a clean API. GLM-5.3 (#9) Modern, Pythonic API for reading and writing PDFs. Gemini 4 Argon (#10) Feature-rich pure Python library supporting both PDF creation and structured reading.
11	Camelot	7/10	9	8			10	10	10	10		7
Nemotron 3.5 Super VL (#7) Best for extracting tables from PDFs. GPT-6 Luna (#8) Useful for extracting tables from text-based PDFs with configurable parsing approaches. Claude Haiku 5.5 (#9) Specialized and reliable for extracting tables from PDFs into dataframes. Kimi K3 (#10) Excellent table extraction with visual debugging via web UI. MiniMax M3 (#10) Specialized for extracting tables from PDFs with high accuracy
12	pdf2image	2/10					9					8
DeepSeek V4.1 Flash (#9) Simple, reliable PDF-to-image conversion via Poppler; ideal for OCR pipelines.
13	Boruta	1/10								7
No reasons recorded.
14	pypdfium2	1/10			7
Gemini 4 Argon (#7) Permissively licensed binding to Chromium's PDF engine for fast rendering and viewing.
15	pdfkit	1/10								8
MiniMax M3 (#8) Simple wrapper around wkhtmltopdf, fast and easy for web to PDF

5 lower-scoring picks are not shown; every ballot below lists them in full.

Most divisive

Picks most models listed, at very different positions

PyPDF2#1 for Nemotron 3.5 Super VL #9 for GPT-6 Luna
PDFMiner.six#2 for Nemotron 3.5 Super VL #9 for Gemini 4 Argon
pypdf#1 for Qwen 3.8 Flash #6 for Nemotron 3.5 Super VL

Hot takes

Picks that appear on only one ballot

pypdfium2 at #7, only Gemini 4 Argon
Boruta at #7, only MiniMax M3
weasyprint at #8, only Qwen 3.8 Flash
pdfkit at #8, only MiniMax M3
Pillow at #8, only Mistral Large 4
OCRmyPDF at #10, only GPT-6 Luna
Borland PDF Library (or commercial: iText / pdftools) at #10, only Mistral Large 4

Every ballot

Hover a pick to find it on the other ballots. Marked picks appear on no other ballot.

Claude Haiku 5.5

ReportLab
pypdf
PyMuPDF
pdfplumber
PDFMiner.six
PyPDF2
WeasyPrint
pikepdf
Camelot
fpdf2

70% with consensus · 52% self-consistent · knowledge cutoff June 2026

GPT-6 Luna

PyMuPDF
pypdf
pdfplumber
ReportLab
borb
WeasyPrint
pikepdf
Camelot
PyPDF2
OCRmyPDF

91% with consensus · 76% self-consistent · knowledge cutoff May 18, 2026

Gemini 4 Argon

PyMuPDF
pdfplumber
pypdf
ReportLab
WeasyPrint
pikepdf
pypdfium2
fpdf2
PDFMiner.six
borb

85% with consensus · 90% self-consistent · knowledge cutoff March 2026

Qwen 3.8 Flash

pypdf
pdfplumber
PyMuPDF
ReportLab
PDFMiner.six
fpdf2
pikepdf
weasyprint
WeasyPrint
borb

72% with consensus · 88% self-consistent · knowledge cutoff not published

DeepSeek V4.1 Flash

pypdf
PyMuPDF
pdfplumber
ReportLab
pikepdf
PDFMiner.six
borb
WeasyPrint
pdf2image
Camelot

76% with consensus · 81% self-consistent · knowledge cutoff not published

GLM-5.3

PyMuPDF
pypdf
pdfplumber
ReportLab
PDFMiner.six
pikepdf
fpdf2
WeasyPrint
borb
Camelot

95% with consensus · 84% self-consistent · knowledge cutoff not published

Kimi K3

PyMuPDF
pypdf
pdfplumber
ReportLab
PDFMiner.six
borb
fpdf2
pikepdf
pdfrw
Camelot

92% with consensus · 71% self-consistent · knowledge cutoff not published

MiniMax M3

ReportLab
pdfplumber
pypdf
PyMuPDF
WeasyPrint
PyPDF2
Boruta
pdfkit
fpdf2
Camelot

56% with consensus · 67% self-consistent · knowledge cutoff not published

Mistral Large 4

PyMuPDF
ReportLab
pdfplumber
PyPDF2
pypdf
PDFMiner.six
WeasyPrint
Pillow
pikepdf
Borland PDF Library (or commercial: iText / pdftools)

79% with consensus · 51% self-consistent · knowledge cutoff not published

Nemotron 3.5 Super VL

PyPDF2
PDFMiner.six
ReportLab
pdfplumber
PyMuPDF
pypdf
Camelot
pdf2image
fpdf2
pdfrw

42% with consensus · 73% self-consistent · knowledge cutoff not published

Who agrees with whom

Overlap between two models' lists, weighted toward the top · 100 means identical · Color runs from 34 (lightest) to 95 (darkest)

	Claude Haiku 5.5	GPT-6 Luna	Gemini 4 Argon	Qwen 3.8 Flash	DeepSeek V4.1 Flash	GLM-5.3	Kimi K3	MiniMax M3	Mistral Large 4	Nemotron 3.5 Super VL
Claude Haiku 5.5		63	55	65	64	67	65	79	66	49
GPT-6 Luna	63		83	64	75	88	88	54	73	35
Gemini 4 Argon	55	83		67	67	85	80	62	73	34
Qwen 3.8 Flash	65	64	67		84	74	72	57	52	40
DeepSeek V4.1 Flash	64	75	67	84		79	76	50	60	38
GLM-5.3	67	88	85	74	79		95	51	75	40
Kimi K3	65	88	80	72	76	95		49	73	41
MiniMax M3	79	54	62	57	50	51	49		60	42
Mistral Large 4	66	73	73	52	60	75	73	60		50
Nemotron 3.5 Super VL	49	35	34	40	38	40	41	42	50