WebP vs JPG, measured: how much smaller WebP really is, and where the gap closes
7 minutes, updated 2026-11-19
WebP is usually described as a quarter to a third smaller than JPG. That figure is an average over many images and settings. It hides two things that matter when you choose a format: the saving depends heavily on what is in the picture, and at the highest quality settings it can turn into a loss. We measured both.
How we measured: on 19 November 2026 we encoded three public-domain NASA photographs, each 1920 pixels wide, and a synthetic 1440 × 900 screenshot of dark text on white, at qualities from 95 down to 30 with Chromium's built-in JPG and WebP encoders. These are the same encoders the tools on this site use in your browser. We also saved each image as PNG. Fidelity is PSNR in decibels (dB) against the image before encoding: above 40 dB is hard to tell apart, 35 to 40 is good, 30 to 35 shows on close inspection, below 30 is obvious. The JPG quality guide explains the measure and lists every JPG figure.
Same quality number
The simplest comparison sets both encoders to the same number. The table gives the WebP file as a percentage of the JPG file at that setting, so anything under 100 means WebP was smaller. The photos are a crew group portrait and an astronaut with two dogs (both 1920 × 1536) and a tall single portrait (1920 × 2400).
| Quality | Crew (% of JPG) | Dogs (% of JPG) | Tall portrait (% of JPG) | Screenshot (% of JPG) |
|---|---|---|---|---|
| 95 | 105 | 105 | 118 | 65 |
| 90 | 92 | 87 | 100 | 64 |
| 85 | 79 | 73 | 77 | 63 |
| 80 | 69 | 65 | 70 | 63 |
| 70 | 80 | 77 | 83 | 64 |
| 60 | 79 | 78 | 77 | 69 |
| 50 | 75 | 76 | 69 | 72 |
| 40 | 80 | 82 | 71 | 76 |
| 30 | 80 | 81 | 67 | 80 |
At 95, WebP was larger than JPG on all three photos, by 5 to 18%. The widest gap on photos, at 80, is partly a JPG quirk: Chromium's JPG encoder stays large at 80 and only drops sharply at 70. The screenshot came out at 63 to 80% of the JPG at every setting.
The same number does not buy the same fidelity, though. At 80, WebP measured 36.1 to 37.4 dB on the photos against 36.4 to 39.2 dB for JPG, so part of its saving was paid for in quality. At 30 the order reversed: WebP 31.2 to 32.2 dB, JPG 29.8 to 32.0 dB. A fair comparison needs equal fidelity.
Equal fidelity
Here each JPG is paired with the WebP setting whose PSNR came closest. The pairs are within 0.7 dB of each other, except the tall portrait at the top level, where even WebP at 95 fell 1.5 dB short of JPG at 85. Each cell reads quality / KB / dB.
| Image and level | JPG (q / KB / dB) | WebP (q / KB / dB) | WebP vs JPG (%) |
|---|---|---|---|
| Crew, high | 85 / 707 / 41.5 | 95 / 978 / 41.8 | +38 |
| Dogs, high | 85 / 543 / 44.6 | 95 / 757 / 44.1 | +39 |
| Tall portrait, high | 85 / 986 / 45.8 | 95 / 1,542 / 44.3 | +56 |
| Crew, good | 80 / 638 / 36.4 | 80 / 438 / 36.1 | −31 |
| Dogs, good | 80 / 490 / 38.8 | 85 / 394 / 39.2 | −20 |
| Tall portrait, good | 80 / 869 / 39.2 | 85 / 759 / 38.5 | −13 |
| Crew, medium | 70 / 394 / 33.1 | 60 / 271 / 33.6 | −31 |
| Dogs, medium | 70 / 302 / 34.9 | 60 / 203 / 34.7 | −33 |
| Tall portrait, medium | 70 / 483 / 34.7 | 70 / 401 / 34.9 | −17 |
| Screenshot, high | 90 / 250 / 41.7 | 80 / 124 / 41.3 | −50 |
| Screenshot, good | 80 / 198 / 37.5 | 50 / 97 / 37.6 | −51 |
| Screenshot, medium | 70 / 170 / 34.7 | 30 / 83 / 35.1 | −51 |
Three results follow from the table.
Photos at screen quality, 33 to 39 dB: WebP was 13 to 33% smaller. The crew and dog photos saved 20 to 33%, the tall portrait 13 to 17%. That is close to the usual claim, with a wide spread from one photo to the next.
Photos at archive quality, above 41 dB: JPG was smaller on all three. To match JPG at 85, WebP needed 95 and produced files 38 to 56% larger, and on the tall portrait it still fell short. If you keep high-quality copies of photos, JPG is the more compact choice.
The screenshot: WebP was about half the size of JPG at every level. WebP's lossy mode comes from the VP8 video codec. It predicts each block from the pixels already decoded around it and can work in smaller blocks, which suits flat areas and hard edges. JPG codes fixed 8 × 8 blocks and spends many bytes on the ripples it leaves around each letter. Photos have few flat areas, so there the advantage is smaller.
At smaller pixel sizes, WebP won on both counts
We also resized each photo to 1280, 800, 600 and 400 pixels wide and saved it at 80 and 60. At every one of those sizes, on all three photos and at both settings, WebP produced the smaller file and measured closer to its source. At 80 the saving ranged from 17 to 37%.
| Width (px) | Crew JPG (KB) | Crew WebP (KB) | Dogs JPG (KB) | Dogs WebP (KB) | Tall JPG (KB) | Tall WebP (KB) |
|---|---|---|---|---|---|---|
| 1280 | 247 | 186 | 188 | 135 | 277 | 196 |
| 800 | 109 | 85 | 80 | 59 | 105 | 66 |
| 600 | 68 | 55 | 48 | 37 | 64 | 41 |
| 400 | 34 | 28 | 25 | 19 | 33 | 23 |
This is the size range of thumbnails, product grids and article images, so on a website the real-world saving is often better than the full-size figures suggest.
Converting JPGs you already have
All three photos started out as JPGs, which is the usual situation when you move a site's images to WebP. Every WebP figure on this page therefore includes the cost of a second encode, and the PSNR is measured against the decoded JPG, the picture you actually hold.
Converted straight from the source files, WebP at 80 came out 40 to 43% smaller than the originals at 36.1 to 37.4 dB. WebP at 85 was 23 to 30% smaller at 37.8 to 39.2 dB. At 70 the files shrank further but fidelity fell to 34.3 to 35.5 dB, the range where a close look finds softening. For photos that people will look at closely, such as product shots, 85 is the safer setting; for supporting images in an article, 80 is enough.
PNG for photos and for screenshots
PNG keeps every pixel exactly. For a photo that meant 8.5 to 9 times the size of a JPG at 85, and 8.2 to 8.7 times the source file. Converting a JPG photo to PNG does not improve it: the file stores the existing JPG artefacts losslessly, at that cost.
For the screenshot, PNG at 214 KB was the same size as JPG at 85 and perfectly exact, which makes it the better of those two for text. WebP at 80 was 42% smaller than the PNG at 41.3 dB.
Our screenshot is a hard case for PNG, because anti-aliased text fills most of the frame. Real screenshots with large empty areas, or ones reduced to a 256-colour palette as the compress tool does for PNG, usually come out much smaller. We did not measure lossless WebP.
| Image | PNG (KB) | JPG at 85 (KB) | WebP at 80 (KB) | PNG vs JPG 85 |
|---|---|---|---|---|
| Crew, 1920 × 1536 | 5,983 | 707 | 438 | 8.5 × |
| Dogs, 1920 × 1536 | 4,903 | 543 | 317 | 9.0 × |
| Tall portrait, 1920 × 2400 | 8,704 | 986 | 607 | 8.8 × |
| Screenshot, 1440 × 900 | 214 | 219 | 124 | 1.0 × |
Where WebP is accepted, and where it is not
A smaller file is no use if it is refused. Government, exam and job portals usually list JPG, sometimes PNG, and check the file's contents, so renaming a WebP does not get it through. Some email clients and older office software do not show WebP either. The WebP guide lists support in detail.
What to use
To change format, use JPG to WebP or WebP to JPG. Both run in your browser and the image is never uploaded.
Before you switch a whole site, try a few of your own pictures. Drop them into the compress tool, choose WebP in the format list, and read the size in each row; Compare shows the result next to the original. Your photos are not ours, and the spread above (13 to 33% on three photos of the same kind) says the only reliable figure is the one you measure.
| Situation | Format and quality | Based on |
|---|---|---|
| Photo on your own website | WebP, 70 to 85 | 13 to 33% smaller than JPG at equal fidelity |
| Thumbnail or grid image | WebP, 80 | 17 to 37% smaller and closer than JPG |
| Screenshot, chart or UI on a website | WebP, 80 to 90 | About half the JPG size; 41 to 45 dB |
| Portal or form upload | JPG | What validators accept |
| Archive copy of a photo | The original JPG | WebP at 95 was larger than JPG at 85 |
| Screenshot where WebP is refused | PNG | Exact, and no bigger than JPG at 85 here |