← All guides

WebP vs JPG: how much smaller, measured on 24 photos

The figure you will find almost everywhere is that WebP is 25 to 34 percent smaller than JPG. It comes from a study Google published years ago, and this site repeated it too. So we measured it with the encoders that actually run when you convert a photo in a browser today. The short answer: about a fifth to a quarter smaller for a photo that looks the same — real, but less than the famous number. And if your photo is already a JPG, converting it can make the file bigger.

What we measured

We took the 24 photos of the Kodak test set, the collection most image compression studies use (Google’s included), each 768 by 512 pixels and stored losslessly. In Chrome 155 we saved every photo as JPG and as WebP at 36 quality settings, from 0.30 to 1.00, using the same browser call our tools use. Then we decoded each file again and scored it against the original with SSIM, a standard measure of how alike two pictures look, where 1 means identical. That is 1,728 encodes, each one scored.

At the same quality number

Image tools ask for a quality between 0 and 1 (or 0 and 100). The same number does not mean the same thing in both formats, but it is what most people compare. Here are all 24 photos together:

Quality settingJPGWebPWebP smaller by
0.92 (our “High”)2,959 KB2,649 KB10%
0.90 (our converter)2,678 KB2,323 KB13%
0.80 (our “Balanced”)1,796 KB1,410 KB21%
0.62 (our “Small”)1,208 KB978 KB19%

On average the WebP files also scored slightly higher on SSIM than the JPGs at the same setting, so across the set the saving is not bought with a worse picture. Photo by photo it is less tidy: at 0.62, half the WebPs scored a little lower than their JPG, and at three of the four settings one photo came out bigger as WebP, by up to 7 percent. The saving is also smallest where people most often set the dial: at 0.92, a tenth.

At the same look

The fairer comparison, and the one Google’s study made, is to match the picture rather than the number: for each JPG, find the smallest WebP that scores at least as well on SSIM.

JPG saved atWebP smaller by (all 24)Range across photos
0.9221%0 – 31%
0.9024%13 – 30%
0.8023%2 – 31%
0.6223%15 – 32%

So: about 21 to 24 percent across the whole set, and never bigger — for every photo at every setting, some WebP matched the JPG at a smaller size. The study’s 25 to 34 percent compared WebP with an old version of the standard JPG encoder (libjpeg 6b); the JPG encoder in today’s browsers is better, which closes part of the gap.

Which photos gain the most — it depends how you compare

The usual advice is that WebP wins on smooth pictures and barely helps on busy ones. At the same quality number that is what we saw. At 0.80 a close-up of two parrots against a soft green background went from 47 KB as JPG to 29 KB as WebP, 38 percent smaller, while a mountain stream full of pebbles, foam and pine needles went from 127 KB to 123 KB, just 3 percent.

Matched on look, it reversed. The parrots’ WebP had been smaller partly because it kept less: at 0.80 it scored lower than the JPG. To score as well it needed 0.88, and the saving fell to 2 percent. The stream’s WebP could go down to 0.60 and still match, and saved 29 percent. So the rule of thumb describes what WebP does at a given setting, not where it is genuinely more efficient — and if you judge by eye on smooth gradients, check the result rather than trusting the number.

Converting a photo that is already a JPG

Most people are not choosing a format for a fresh photo. They have a JPG and want it smaller, so they convert it to WebP. That is a second lossy save on top of the first, and the result depends on how hard the JPG was compressed to begin with. We saved each photo as a JPG, then converted that JPG to WebP at 0.90 — what our JPG to WebP converter does:

Original JPG saved atWebP at 0.90 (all 24)Photos that got bigger
0.9538% smaller0 of 24
0.9222% smaller0 of 24
0.853% bigger14 of 24
0.8019% bigger24 of 24

A JPG saved at high quality shrinks a lot. A JPG that was already squeezed grows, because the converter is spending more bits than the original did, on a picture that has already lost detail. Every conversion also lowered the SSIM score a little, as a second lossy save always does. If the file only needs to be smaller, the image compressor is the better tool: it starts from your file and refuses to hand back something bigger.

Quality 1.0 is a special case

In Chrome, WebP at quality 1.0 switches to lossless: every one of the 24 photos decoded back pixel-for-pixel (SSIM exactly 1). Those files were 42 percent smaller than the same photos as PNG — but nearly three times the size of WebP at 0.98, which already looks the same to the eye. JPG has no lossless mode; at 1.0 it still scored just under 1. Unless you need the exact pixels, 1.0 is the wrong setting.

What we did not measure

The test photos are small by today’s standards: about 0.4 megapixels, scanned from film. A 12-megapixel phone photo has different noise and detail, and the percentages could move. We measured only Chrome’s encoders; Safari and Firefox ship their own, and they can differ. And SSIM, like any score, is a stand-in for your eyes, measured here on brightness only. Treat the numbers as what one browser does with one well-known set of photos, not as a law.

So which should you use?

More guides

All guides →