Original test data
We tested our tools. Here are the numbers.
Exact results from running Compress PDF and the Image Compressor on public test files on 16 October 2026 β including the file that did not get smaller.
Key findings
What we found
β75.9%
A 10-page colour scan with Compress PDF on Recommended (0.9 s).
β62.2%
24 JPG photos with the Image Compressor's default settings, at 35.7 dB PSNR.
0%
A text-only government PDF in the browser β nothing safe to remove, so it was returned unchanged rather than made bigger. Our server engine cut the same file by 57%.
Compress PDF
PDF compression results
We tested it: real results
Measured by MyToolSpace on 16 October 2026 with this tool's own engine and public test files β exact numbers, not estimates.
Text-heavy government PDF (IRS Form 1040 instructions, 126 pages)
Original: 4.23 MB
No preset could make this file smaller β it holds mostly text and fonts, with no images that can be safely recompressed. The tool returned the original unchanged instead of a larger file.
Photo report: 24 photos saved as JPEG (quality 95), one per A4 page
Original: 3.73 MB
| Setting | Result | Saved | Time |
|---|---|---|---|
| Low compression (best quality) | 2.09 MB | β44.1% | 0.7 s |
| Recommended | 1.53 MB | β58.9% | 0.6 s |
| Extreme (smallest file) | 1.16 MB | β69% | 0.2 s |
Colour scan: 10 pages at 300 DPI saved as JPEG page images
Original: 9.28 MB
| Setting | Result | Saved | Time |
|---|---|---|---|
| Low compression (best quality) | 4.45 MB | β52% | 2.7 s |
| Recommended | 2.24 MB | β75.9% | 0.9 s |
| Extreme (smallest file) | 988 KB | β89.6% | 0.7 s |
Every output kept all of its pages; on the text document, all selectable text was preserved.
Large files: server engine (Ghostscript)
Files over 50 MB are compressed on our servers. The same documents, plus a 70-page scan, through that engine:
| Document | Original | Recommended | Extreme | Time |
|---|---|---|---|---|
| Text-heavy government PDF (IRS Form 1040 instructions, 126 pages) (126 pages) | 4.23 MB | 1.82 MB (β57%) | 1.71 MB (β59.5%) | 6.4 s |
| Photo report (24 pages) | 3.73 MB | 1.22 MB (β67.3%) | 744 KB (β80.5%) | 0.4 s |
| Colour scan (10 pages) | 9.28 MB | 1.95 MB (β79%) | 806 KB (β91.5%) | 1 s |
| Large colour scan (70 pages) | 78.98 MB | 16.19 MB (β79.5%) | 6.35 MB (β92%) | 6.2 s |
Every server output kept all of its pages. On the text document, 13 of 496,923 characters (0.003%) copy or search differently afterwards β for example an em dash (β) β because Ghostscript re-encodes fonts. The pages look the same; the browser engine changed none.
Test machine: 13th Gen Intel(R) Core(TM) i7-13620H, 16 threads, 24 GB RAM, Chromium 131.0.6778.33 (headless); server engine: Node.js v25.9.0, Ghostscript 10.08.0. Your results depend on what is inside your file. How we tested
Image Compressor
Image compression results
We tested it: real results
Measured by MyToolSpace on 16 October 2026 with this tool's own engine and public test files β exact numbers, not estimates.
24 photos as PNG (768Γ512, Kodak test suite)
Original total: 14.68 MB for 24 images
| Setting | Output | Result | Saved | Quality (PSNR) |
|---|---|---|---|---|
| Default settings (Smart mode, quality 80) | WEBP | 1.38 MB | β90.6% | 36.4 dB |
| Keep as JPEG, quality 80 | JPG | 1.75 MB | β88.1% | 35.4 dB |
The same 24 photos as JPEG, quality 95 (like a camera export)
Original total: 3.72 MB for 24 images
| Setting | Output | Result | Saved | Quality (PSNR) |
|---|---|---|---|---|
| Default settings (Smart mode, quality 80) | WEBP | 1.41 MB | β62.2% | 35.7 dB |
| Keep as JPEG, quality 80 | JPG | 1.80 MB | β51.7% | 34.9 dB |
PSNR compares each result with the original photo: above 40 dB is visually identical for most people, 35β40 dB is very good.
Test machine: 13th Gen Intel(R) Core(TM) i7-13620H, 16 threads, 24 GB RAM, Chromium 131.0.6778.33 (headless); server engine: Node.js v25.9.0, Ghostscript 10.08.0. Your results depend on what is inside your file. How we tested
Method
How we tested
Each tool's own engine was run in the browser exactly as the tool page runs it (Compress PDF in a Web Worker). Sizes are exact byte counts; times are the median of 3 runs for PDFs and per image for photos. Image quality is PSNR against the original photo (above 40 dB is visually indistinguishable for most people; 35β40 dB is very good).
Test machine: 13th Gen Intel(R) Core(TM) i7-13620H, 16 threads, 24 GB RAM, Chromium 131.0.6778.33 (headless); server engine: Node.js v25.9.0, Ghostscript 10.08.0.
Test files (public, so anyone can repeat the test):
- IRS Instructions for Form 1040 and 1040-SR β US government work, public domain.
- Kodak Lossless True Color Image Suite (24 images) β Released by Kodak for unrestricted usage.
- The photo report puts the 24 Kodak photos, saved as JPEG at quality 95, one per A4 page.
- The colour scan is the first 10 pages of the IRS document rendered at 300 DPI and saved as JPEG page images, like a scanner's output.
For the PDFs we also checked that every output kept all of its pages and, for the text document, all of its selectable text.
Server engine: the same Compress PDF server code that handles files over 50 MB (Ghostscript followed by our lossless pass), run once per setting on the test machine. Our production servers have different hardware, so their times will differ.
Last updated
Questions about these tests
Are these numbers estimates?
No. Every size is the exact byte count of the file the tool produced, and every time was measured on the test machine. The script that produced them is part of our codebase, so the test can be rerun at any time.
Why did the text PDF not get smaller in the browser?
That document is almost entirely text and embedded fonts; its only images use CMYK colour, which the browser engine leaves untouched to avoid colour shifts. There was nothing to remove safely, so Compress PDF returned the original file instead of a bigger one. The server engine (Ghostscript), used for files over 50 MB, re-encodes fonts and CMYK images and cut the same file by 57% on Recommended.
Will my files compress by the same amount?
Only if they look like the test files. Scans and photo-heavy PDFs shrink a lot because their images can be downsampled; text-only PDFs and already-compressed files shrink little or not at all.
What does PSNR mean?
Peak signal-to-noise ratio compares the compressed image with the original pixel by pixel. Above 40 dB the difference is invisible to most people; 35β40 dB is very good quality; below 30 dB artefacts become noticeable.
Did you test the server engine too?
Yes. The Image Compressor always runs in your browser, and Compress PDF does too for files up to 50 MB, so those were tested in a browser (Chromium 131.0.6778.33 (headless)). PDFs over 50 MB are compressed on our servers with Ghostscript; we ran that same engine on the same documents plus a 70-page scan of 78.98 MB.