Merging and Splitting PDFs in Your Browser: Real Times
By Junaid Aslam. 6 min read. Updated .
The PDF Tools page merges and splits PDFs in the browser. It reads the file into memory, runs pdf-lib on the page, and builds a download blob. No PDF goes to a server. That design sets the real limits: the file size the page accepts, the memory the tab has, and the time each operation takes.
We put three current NIST PDFs through the page, three runs each. Merging two files took 1.37 to 1.42 seconds. Splitting took 0.36 to 1.46 seconds. The split output was much larger than the input, and the page dropped a 51MB file without a message.
The three test files
| File | Title | Pages | Bytes |
|---|---|---|---|
| FIPS 180-4 | Secure Hash Standard (SHS), August 2015 | 36 | 833,315 |
| FIPS 197 | Advanced Encryption Standard (AES), updated 9 May 2023 | 46 | 1,184,436 |
| FIPS 186-5 | Digital Signature Standard (DSS), 3 February 2023 | 86 | 1,844,414 |
All three are current NIST publications, not drafts or withdrawn files. NIST says information on its site is public information and may be distributed or copied, except material marked as copyrighted. We searched the text of all three files and found no page with a copyright notice.
How the runs were made
The test ran on 2026-11-17 against commit 357b336 on a local production build (Next.js 14.2.35). The machine is a Windows 10 desktop with an Intel Core i7-1165G7 (4 cores, 8 threads) and 15.7 GiB of RAM. The browser was Chromium 159.0.8154.6, driven by Playwright 1.65.1 on Node 24.20.0.
Each run opened a new browser context. We uploaded the files, waited for the page count and the enabled action button, then started the clock and clicked. For merge, the clock stopped when the Download button appeared, which means the merged blob was ready. For split, it stopped when the Split Pages (N) heading appeared, which means every page blob was built. Playwright polls the page, so each time can include up to about 50 ms of polling delay.
We saved the merged PDFs and checked their page counts with pypdf. For split, we counted the page links and recorded the size of every blob the page created.
Merge results
| Files | Pages out | Input bytes | Output bytes | Runs (ms) | Min | Max |
|---|---|---|---|---|---|---|
| FIPS 180-4 + FIPS 197 | 82 | 2,017,751 | 1,514,308 | 1389, 1390, 1418 | 1389 | 1418 |
| FIPS 197 + FIPS 186-5 | 132 | 3,028,850 | 2,625,682 | 1398, 1384, 1374 | 1374 | 1398 |
| All three | 168 | 3,862,165 | 3,236,689 | 1407, 1966, 1394 | 1394 | 1966 |
The output page counts match the input totals: 36 plus 46 is 82, 46 plus 86 is 132, and all three is 168. The merged file was 13 to 25 percent smaller than the sum of its inputs. The page copies each input page into a new document, so objects the inputs do not use do not carry over.
The all-three set had one slower run at 1,966 ms. The other two runs were 1,407 and 1,394 ms. That is a real spread, not a typo, and it is why the table shows all three times.
Split results
| File | Pages | Input bytes | Output bytes | Runs (ms) | Min | Max | Bytes per page |
|---|---|---|---|---|---|---|---|
| FIPS 180-4 | 36 | 833,315 | 2,377,686 | 989, 360, 364 | 360 | 989 | 66,047 |
| FIPS 197 | 46 | 1,184,436 | 4,605,014 | 1462, 1378, 1425 | 1378 | 1462 | 100,109 |
| FIPS 186-5 | 86 | 1,844,414 | 24,689,782 | 887, 877, 882 | 877 | 887 | 287,090 |
Every split produced one file per page, so the page counts are 36, 46, and 86. The total output is much larger than the input: 2.4MB from 0.83MB, 4.6MB from 1.18MB, and 24.7MB from 1.84MB. The 86-page file produced 13.4 times its input size. Each page becomes a new PDF, and shared resources are written again for each one.
The 46-page AES file took longer to split than the 86-page DSS file. Page count alone did not predict the time in this test. The AES file has more diagrams and images per page, and those take more work to copy.
The first split of the 36-page file took 989 ms. The next two took 360 and 364 ms. The first run paid a warm-up cost for the PDF engine in that browser session. A single run would have overstated the normal time by about 2.7 times.
The file-size limit we hit
Both uploaders pass maxSize={50 * 1024 * 1024} to the shared FileUpload component. The dropzone prints Max size: 50MB under the button. The component keeps a file only when file.size <= maxSize.
We tested the boundary with two padded files. A 49MB file appeared as Selected Files (1). A 51MB file produced no selected file, no visible error, and no explanation. The page did not crash. The uploader dropped it before reading it. The cap is per file, not per merge, and a 50MB file is the largest that passes.
This is a file-size limit, not a memory limit. We did not hit a memory error in any run. The largest test set was 3.86MB across three files, far below the cap. We did not test a 50MB merge or a 50MB split, so this guide does not quote a memory ceiling.
The FAQ on the same page says files over about 100MB can be slower. In this build, the uploader drops any file over 50MB before it is read. That mismatch is worth knowing before you try a large scan.
What this means in practice
- For files under a few megabytes, merge and split finish in about one to two seconds on this machine.
- Split output can be much larger than the input. An 86-page, 1.84MB file became 24.7MB of page files. Check free disk space before you download all of them.
- Merge output can be smaller than the inputs. In this test it was 13 to 25 percent smaller.
- A 51MB file is dropped without a message. If a file does not appear in the list, check its size first.
- One run is not enough to judge speed. The first split run was 2.7 times slower than the next two, and one merge run was 41 percent slower than the other two.
Tested on
2026-11-17, against commit 357b336 on a local production build (Next.js 14.2.35), with pdf-lib 1.17.1 in the page. Browser: Chromium 159.0.8154.6 through Playwright 1.65.1. Host: Windows 10, Intel Core i7-1165G7 (4 cores, 8 threads), 15.7 GiB RAM, Node 24.20.0. Three runs per merge set and per split file.
Primary sources
- NIST: Copyrights and Disclaimers: the statement that NIST site information is public information and may be distributed or copied, except material marked as copyrighted.
- FIPS 180-4: Secure Hash Standard: the 36-page test file.
- FIPS 197: Advanced Encryption Standard: the 46-page test file.
- FIPS 186-5: Digital Signature Standard: the 86-page test file.
- pdf-lib: the PDF engine the page loads.
Document sources and photo credits
This guide has no photos. The three test files are NIST publications:
- FIPS 180-4, Secure Hash Standard (SHS), August 2015, 36 pages, 833,315 bytes: PDF.
- FIPS 197, Advanced Encryption Standard (AES), updated 9 May 2023, 46 pages, 1,184,436 bytes: PDF.
- FIPS 186-5, Digital Signature Standard (DSS), 3 February 2023, 86 pages, 1,844,414 bytes: PDF.
Related tools
- PDF Tools: merge or split a PDF in your browser.
- Image Converter: convert images without an upload.
- CSV Cleaner: clean a spreadsheet in the browser.
- JSON Formatter: format or query JSON locally.