Benchmark: how small, at what quality

What PicPinch's engine does to 204 test images, how it compares with TinyPNG, and how we measured it. Measured in November 2026; the website's engine makes byte-identical files.

In short

  • Against TinyPNG, at the same quality: 49% of TinyPNG's total size with WebP, 59% with PNG and JPEG only · 18 PNG images
  • Against the originals, at the default quality: 36% of their total size, 27% with WebP · 204 images
  • Every file checked: each result is decoded again and scored; none is below its target, and its most-changed area is at most 5 points lower

Against TinyPNG

We compressed 18 PNG images (8 screenshots, 2 diagrams, 8 photos) on TinyPNG's website, scored each of TinyPNG's files, and asked our engine for the same score.

Total size, as a share of TinyPNG'sMedian imageImages bigger than TinyPNG's
PNG and JPEG only59.0%87.4%2 (by 19% and 1%)
With WebP49.4%56.9%0

Photos came out at 16–55% of TinyPNG's size, because TinyPNG kept them as PNG while our engine saved them as JPEG or WebP. Screenshots and diagrams kept as PNG were about even (82–119%), and 54–96% with WebP.

Against the original files

All 204 test images, at each quality level. A file that can't get smaller is kept as it is.

QualityTotal size, as a share of the originals: PNG and JPEG / with WebPImages made smaller: PNG and JPEG / with WebP
Standard (80)35.6% / 26.6%199 / 204
High (85)47.5% / 35.3%188 / 202
Highest (90)75.8% / 58.8%104 / 168

The test images

204 images: 155 from a public compression test set (34 3D renders, 30 icons, 39 illustrations, 24 photos and 28 photos with transparency); the 18 used against TinyPNG (their 8 photos are the same Kodak images as in that set, so they count twice); 15 camera JPEGs of about 10 megapixels; and 16 app and website screenshots.

How quality is measured

With SSIMULACRA2, a 0–100 measure built to match what people see. 80 is our default (looks the same at normal size), 85 is High and 90 is Highest. Every output file is decoded again and scored against its original, as a whole and tile by tile. The scorer is the same one our engine aims for, which favours it slightly.

Speed in the browser

On our test Mac, in headless Chromium and WebKit with 4 threads, our 30-image sample took a median of about 2 seconds per image, and 11 seconds for the slowest. Phones use one thread and are slower; we haven't measured them on real devices yet.

Limits of this test

  • The TinyPNG comparison has only 18 images, all PNG
  • The scorer is the one our engine optimises for
  • TinyPNG's speed wasn't measured: its website refused automated uploads, and we didn't work around that
  • Phones and Firefox are not measured yet

Questions

Is PicPinch better than TinyPNG?
In our test it made smaller files at the same quality: 59% of TinyPNG's total size with PNG and JPEG, 49% with WebP. TinyPNG has an API and plugins, which PicPinch doesn't have yet.
What does a score of 80 mean?
On the SSIMULACRA2 scale, 80 means you won't see a difference at normal size; differences show only when you zoom in a lot. 90 is indistinguishable even up close.

Related tools