WebP vs JPG: What the 25-35% Saving Actually Looks Like
"WebP is 25-35% smaller than JPEG" is the sentence that sells the format, and it is broadly true. It is also an average across a benchmark corpus, and the variance underneath it is large enough that the average is a poor guide to any particular image.
Where WebP's advantage is largest
Images with flat areas and gradients. Skies, studio backdrops, UI screenshots with photographic content, product shots on white. JPEG's 8x8 block transform produces visible banding and blocking in smooth gradients unless you spend bytes on it. WebP handles them substantially better, and the size gap can exceed 40%.
Anything with transparency. This is not a percentage difference, it is a categorical one. JPEG has no alpha channel at all, so a logo with a transparent background has to be a PNG — which for anything photographic means an enormous file. WebP does lossy compression with full 8-bit alpha, which no other mainstream format offers. Here the saving is routinely 80% or more.
Low-quality thumbnails. At aggressive settings WebP holds together noticeably better. Where a JPEG at quality 40 is visibly falling apart, a WebP at the same file size still looks acceptable.
Where the advantage narrows
Highly detailed, high-noise photographs. Dense foliage, textured fabric, film grain, high-ISO noise. Both codecs have to spend bits describing genuinely unpredictable detail, and WebP's edge shrinks to perhaps 10-15%.
Already-compressed sources. Converting a JPEG that was saved at quality 60 gets you a WebP of a degraded image. The artefacts are now part of the picture and the encoder faithfully preserves them. Always convert from the highest-quality source available.
Very small images. Below about 10 KB, container overhead starts to matter and the percentages become noise.
The quality scales are not comparable
A frequent source of bad benchmarks: comparing WebP quality 80 against JPEG quality 80 and concluding something about the formats. The numbers are encoder-specific and measure different things. In practice a WebP at 75 tends to look closer to a JPEG at 85 while being smaller.
The only honest comparison is at matched visual quality: encode both, view them at the size you will actually display, and compare file sizes at the point where you cannot tell them apart.
When JPG is still the right answer
Compatibility, and it is a bigger deal than performance writing tends to admit. Every device, editor, printer, email client and upload form ever built accepts a JPG. WebP works flawlessly in browsers and unevenly outside them — Photoshop only from 23.2, Office only from 2021, Outlook not at all inline.
The practical division: serve WebP on your website, hand out JPG when a file is going to a person, a printer, an email or a form. Most teams do both via a <picture> element, which costs one extra build step and removes the need to choose.
What to do with this
Do not take the 30% figure on faith and do not dismiss it either. Take ten representative images from your own site, convert them, and measure. Ten minutes of real numbers from your own content beats any benchmark article, including this one.