The median page, in two numbers
| Requests | Weight | |
|---|---|---|
| 25th percentile | 69 | 1.7 MB |
| Median | 102 | 3.4 MB |
| 75th percentile | 171 | 6.2 MB |
| 90th percentile | 235 | 11.8 MB |
| Maximum | 300 (capped) | 72.9 MB |
For scale: the HTTP Archive measured the median mobile page at 2,311 KB in October 2024. Ours is about a megabyte heavier, and that gap is the sample rather than the year. Sites arrive in our base because somebody wanted to know why they were slow.
What the weight is made of
Each row below covers only the sites that carry that resource type, so the medians describe a typical page with video, or with fonts, rather than a blended average of pages that have none.
| Type | Sites | Median weight | 90th pct | Median share of the page |
|---|---|---|---|---|
| Script | 447 | 1.1 MB | 2.6 MB | 35% |
| Image | 448 | 771 KB | 5.4 MB | 28% |
| Media (video) | 90 | 3.2 MB | 18.6 MB | 43% |
| Font | 444 | 209 KB | 548 KB | 6% |
| Stylesheet | 430 | 86 KB | 229 KB | 3% |
| Document (the HTML) | 457 | 59 KB | 191 KB | 2% |
Two rows are worth stopping on.
Images have the wider spread by far. Their median is 771 KB, well under JavaScript’s, but their 90th percentile is 5.4 MB — seven times their own median, against two and a half times for scripts. Image weight is a problem that a minority of sites have very badly, while script weight is a problem almost everyone has moderately.
Video is the sleeper. It appears on 90 sites, one in five, and where it appears it is usually the largest thing on the page: a median of 3.2 MB and 43% of total weight, rising to 18.6 MB on the heaviest tenth. Twenty per cent of a sample is easy to average away, and this is the row least visible in any figure that blends all sites together.
Which type is heaviest, per site
Averages across a sample cannot answer “what should I fix first” for one site. Counting, per site, which type carries the most bytes can:
| Heaviest type on the page | Sites | Share |
|---|---|---|
| Script | 205 | 45% |
| Image | 174 | 38% |
| Media | 47 | 10% |
| Font | 16 | 4% |
| Everything else | 15 | 3% |
JavaScript takes first place on more sites than images do. It is a close-run thing rather than a rout — and it goes the same direction the HTTP Archive reported in 2024, when its own new analysis found that images are not always the biggest component of page weight and that JavaScript overtook them on inner pages.
The stronger version of the point is in the tails: images account for more than half the weight on 132 sites, and scripts do on 128. Roughly three sites in ten are dominated by each, which means the default advice is right about as often as it is wrong.
Sixteen sites are led by their fonts. That is a small group and an entirely avoidable one: 54 sites in the sample ship more than 500 KB of font files, and font weight at that scale is almost always several full families loaded to use two weights of one.
The part that is already solved
215 of the 446 sites with images — 48% — already deliver most of their image bytes as WebP or AVIF. That is the single most encouraging number in this study, and it is worth stating plainly because it contradicts the shape of most performance advice, which still treats modern image formats as the obvious untaken win. On half the sample it has been taken.
Where it has not, the reason is rarely ignorance of the format. It is that the images were uploaded years ago through a CMS that stores what it is given, and nothing since has gone back over them.
What we take from it
The inventory is the reason our own filters are ordered the way they are, and it settled one question internally that argument had not. A generic optimizer has to decide what to spend its effort on before it knows anything about a specific site; the distribution above says that effort belongs on delivery and execution order — what blocks the first paint, what runs during it, what is fetched before it is needed — rather than on squeezing image bytes, which half the sample has already done and a third do not need.
The exception is the 10% carrying video, and it is a genuine exception: no amount of script deferral matters on a page where an autoplaying file is 43% of the weight. Those sites need a different first move, and the honest thing is to tell them so rather than run the standard list at them.
Method and limits
457 audited sites, each measured once in Lighthouse's mobile configuration, inventoried before any of our own optimizations are applied. For every request we record the host, the resource type Lighthouse assigned, the MIME type and the transferred size, then total them per site. Percentiles are across sites, not across requests, so one enormous site cannot drag a median; per-type figures are computed only over the sites that carry that type at all, so the 90 sites with video are not averaged against the 367 without.
Measured on 4 Aug 2026 across 457 sites. These figures are frozen at that date — we don't quietly restate a published study when the audit base grows.
What this doesn't show
- Not a random crawl. 457 sites we chose to audit, weighted toward sites somebody suspected were slow. That shows up directly in the comparison with the HTTP Archive below, where our median page is about 1 MB heavier than the web's, and it is the reason we do not present any of this as a state of the web.
- Our inventory stops at 300 requests per page. 24 sites reached that ceiling, so their totals are floors. This compresses the top of every distribution here, including the 72.9 MB maximum.
- Sizes are transferred bytes as the browser reports them, so a compressed text file counts at its compressed size while an already-compressed image counts at its full size. That comparison is fair to the network and unfair to text: a 1.1 MB median for JavaScript is what crossed the wire, not what the CPU then had to parse.
- One page per site, almost always the homepage, loaded once without interaction. Anything a visitor's click would pull in — a lightbox gallery, a video that starts on tap, a chat widget — is missing.
- Resource types come from Lighthouse's own classification. "Media" is chiefly video but not exclusively; a font delivered with an unusual MIME type can land in "Other". We report the classification as given rather than re-deriving it, so it is consistent, not perfect.
Sources
- Page Weight — Web Almanac, HTTP Archive
The median page weight for a desktop page, as measured in October 2024 is 2,652 KB
- Page Weight — Web Almanac, HTTP Archive
This new analysis also shows that images are not always the biggest component of page weight, as previously thought, and that for inner pages JavaScript took that dubious honor instead.