Get the data out of a PDF
Converting a PDF changes its format; extracting takes the information out of it. The two are usually filed together and they are not the same job - one hands you back a document, the other hands you data. It is also the job where quality varies most, because every extractor manages a clean, ruled table and almost none manages a messy one.
43 tools here do less than the name of this job suggests, and each one's profile says what it actually does. That sentence is on the profile rather than repeated on every row.
43 options · Sources read 18 October 2026 · We have not run the same document through every extractor here, so this page does not rank accuracy. What each tool outputs and how it runs is recorded on its profile.
- What it takes, and what it produces
- In: a PDF. Out: text, tables or structured data.
- What limits it
- A scan needs a text layer before extraction can find anything, so on a scanned input the limit is the file rather than a quota. Page counts are capped per tool where they are capped at all.
Extract tools: controls, file handling and cost
These are different ways to do the same job. The order does not indicate quality, and where we have not checked something the row says so rather than guessing.
- 43
- tools handle this job
- 29
- keep the file on your device
- Observed by us, or no server anywhere in the path. Of the other 14: 6 upload your file; 5 process some versions on your device and run others on a server; 2 are documented as staying local but have not been tested; 1 runs on a server you provide.
43 tools for extracting
Ordered by relevance to this job, not by payment. Our overall ranking is in the full index.
-
Our pick 5: A Go library and CLI for validation, encryption and assembly. Local, and honest enough to say its PDF 2.0 validation is basic.
- Price Free tierLicence Not verifiedYour files Uploads - claimed
Every listing here is a real product with a real website, checked for whether it is still working. Nothing on this page is paid for and no placement can be bought. Where we have not verified something, the row says so rather than guessing.
Choosing between tidy tables and the awkward ones
-
Ruled tables and tidy documents
Almost every extractor manages a table with clear rules and consistent columns. If that is what you have, the choice between tools matters much less than it appears to.
-
Merged cells, multi-line rows and no rules
This is where extractors separate. The failure is quiet: a merged cell becomes two rows, or a wrapped line becomes a row of its own, and the numbers still look plausible.
-
Whether you want data or a document
CSV and JSON give you values; Markdown and HTML keep the shape of the page. Choosing the wrong one means reconstructing the other by hand.
Check the result
- - Sample five cells from different parts of the document and compare them with the page by eye. Errors concentrate in exactly the places you would not think to look.
- - Check a merged cell and a wrapped line specifically, because those are the two structures that most often produce a plausible wrong answer.
- - Check whether empty cells came back as blanks or as zeros.
- - Check numbers, units and footnotes, which may arrive in the wrong column or attached to the wrong value.
Sources and scope
Every fact on this page comes from a vendor's own page. Each profile records the URL we read, the date, and the SHA-256 of the text, so a reader can check the bytes.
- Adobe PDF Extract API 2 sourced claims, read 22 September 2026
- Apache Tika 2 sourced claims, read 1 October 2026
- Azure AI Document Intelligence 2 sourced claims, read 22 September 2026
- Camelot 4 sourced claims, read 22 September 2026
- Marker 2 sourced claims, read 22 September 2026
- PyMuPDF 3 sourced claims, read 22 September 2026
- Unstructured 2 sourced claims, read 16 October 2026
- Amazon Textract 2 sourced claims, read 22 September 2026
- GROBID 2 sourced claims, read 1 October 2026
- LlamaParse 2 sourced claims, read 1 October 2026
- Reducto 2 sourced claims, read 16 October 2026
- Surya 2 sourced claims, read 16 October 2026
- Tabula 2 sourced claims, read 21 September 2026
- Mindee 1 sourced claim, read 16 October 2026
- Apache PDFBox 1 sourced claim, read 22 September 2026
- Nougat 2 sourced claims, read 1 October 2026
- pdfminer.six 2 sourced claims, read 1 October 2026
- Chunkr 4 sourced claims, read 22 September 2026
- Docling 2 sourced claims, read 22 September 2026
- No Fuss PDF 5 sourced claims, read 29 September 2026
- pdfplumber 2 sourced claims, read 22 September 2026
- Poppler 2 sourced claims, read 25 September 2026
- Syncfusion PDF 2 sourced claims, read 22 September 2026
- JPedal 2 sourced claims, read 1 October 2026
- LEADTOOLS PDF SDK 3 sourced claims, read 16 October 2026
- Spire.PDF 2 sourced claims, read 1 October 2026
- PrizmDoc 3 sourced claims, read 22 September 2026
- GroupDocs 2 sourced claims, read 1 October 2026
- PDF-XChange Core API SDK 3 sourced claims, read 22 September 2026
- pypdf 1 sourced claim, read 1 October 2026
- SnackPDF 3 sourced claims, read 22 September 2026
- Telerik Document Processing Libraries 3 sourced claims, read 22 September 2026
- Qoppa Java PDF SDK 2 sourced claims, read 16 October 2026
- SetaPDF-Core 2 sourced claims, read 16 October 2026
- ICEpdf 2 sourced claims, read 18 October 2026
- pdfcpu 1 sourced claim, read 16 October 2026
- Adobe PDF Library 2 sourced claims, read 1 October 2026
- Google Cloud Vision API and Document AI 2 sourced claims, read 22 September 2026
- iText 2 sourced claims, read 22 September 2026
- MuPDF.js 3 sourced claims, read 22 September 2026
- pdfRest 2 sourced claims, read 1 October 2026
- Scanbot SDK 2 sourced claims, read 22 September 2026
Found a changed feature or an incorrect detail? Report a correction.