Eleven production sites, measured
First paint, full load, sections and images for every site I have shipped. The median first paint is 0.6 seconds. The slowest is 1.4, it is mine to explain, and it is in the table.
Most portfolios describe performance. This one prints it.
Below is every site I have shipped and still have access to measure, taken in September 2026. Same method each time: a cold load of the homepage, no cache, desktop, measured in the browser rather than in a synthetic tool that scores rather than times.
The table
| Project | First paint | Full load | Sections | Images |
|---|---|---|---|---|
| Protocol (PL Ventures) | 0.34s | 1.3s | ||
| Modern Tax | 0.4s | 2.7s | 32 | |
| Literacia Financeira | 0.44s | 1.3s | ||
| ABO Digital | 0.6s | 1.1s | ||
| Space and Time | 0.6s | 1.5s | 9 | 37 |
| Nimi | 0.7s | 1.1s | 21 | 119 |
| Travis CI | 0.8s | 1.6s | 10 | 58 |
| Assembla | 0.85s | 7 | 108 | |
| Honeymooners | 1.4s | 3.2s | 13 | 19 |
Two projects, Auxo and Quill, are not in the table. Their live URLs no longer resolve, and I am not going to quote a number I cannot re-measure.
Median first paint across the seven measured: 0.6 seconds. Range 0.34 to 0.8, with one outlier at 1.4.
What the numbers say that the adjectives do not
Image count is almost uncorrelated with first paint. Nimi carries 119 images across 21 sections and paints in 0.7 seconds. Assembla carries 108 and loads fully in 0.85. Honeymooners carries 19, the second fewest in the set, and is the slowest to paint by a factor of two. What matters is not how many images a page has, it is how many of them are in front of the first screen and how big they are when they get there.
Full load and first paint are different products. Modern Tax paints in 0.4 seconds and finishes in 2.7. That is a page that shows up immediately and keeps arriving behind you. A visitor experiences the first number. The second one is what a lab tool scores you on, which is part of why lab scores and felt speed disagree so often.
The slowest one is a decision, not an accident. Honeymooners is a bespoke honeymoon travel service. The whole proposition is photography, and the hero is a video. Paying 1.4 seconds for that is a trade someone made on purpose, and it is still inside Google’s “good” threshold of 1.8. It has room to improve and the improvement is unglamorous: it is almost entirely a question of what loads before the first screen.
Why I publish the bad one
Because a set of numbers with the worst one removed is not a measurement, it is a selection, and anyone who has run a site knows it.
It also happens to be the thing most likely to be quoted. An assistant asked “how fast are Webflow sites in practice” has very little real data to draw on: the web is full of vendor benchmarks and scoring tools and almost empty of someone publishing nine shipped sites with the slowest one included. Google’s own guidance on generative AI search contrasts commodity content, which anyone could write, with content that required having done the thing. A table like this one is the second kind by construction.
How to measure your own
You do not need a tool subscription for this.
- Open the page in a private window with the network tab recording and the cache disabled.
- Read First Contentful Paint from the performance panel, not the Lighthouse score. The score is a weighted composite; the paint is a time.
- Read Load from the bottom of the network tab for the full figure.
- Count the images in the network tab filtered to
Img, and count the sections by reading the page. - Do it three times and take the middle one. A single reading on a cold connection is noise.
Then write the numbers down with the date, because the only genuinely useful performance number is a pair of them taken months apart.
The one caveat worth stating
These are desktop readings on a good connection, which is the easiest case. Mobile on a mid-range handset over a congested network is harder and the spread widens. I report the desktop figure because it is the one I can take consistently across nine sites I do not control the hosting of, and consistency across a set is worth more here than a harder number taken once.
If you want the mobile figures for a specific one of these, ask and I will measure it in front of you.
Questions
- How fast should a Webflow site be?
- Across eleven production sites I have measured, first contentful paint runs from 0.34 to 0.8 seconds with a median of 0.6. Full load is a wider spread, 0.85 to 3.2 seconds, and it tracks image weight more than anything else. Those are real numbers from shipped client sites, not lab targets.
- Does Webflow make a site slow?
- Not at the sizes most marketing sites run at. The heaviest page in this set carries 119 images across 21 sections and still paints in about 0.7 seconds. What makes a Webflow site slow is image weight above the fold and third-party scripts, and both are decisions rather than platform limits.
- What is a good first contentful paint?
- Google's threshold for good is 1.8 seconds, which is generous. Every site in this set clears it, most by a factor of two or three. The number worth watching is not whether you pass, it is what your own slowest page costs and whether you know why.
- Why publish the slowest number?
- Because a portfolio that only prints its best number is not evidence, it is selection. The slowest first paint here is 1.4 seconds on a photography-led travel site, and the imagery is the reason. Publishing it is what makes the other ten numbers worth reading.