the **spiciest** web performance leaderboard would rank saas company home pages making “blazingly fast”-esque claims
put those INP and TBT scores in big bold letters
It should be fair to post screenshots like this for companies that tout “the fastest frontends” on their home page—especially notable that this is *not* a temporary regression. It’s been like this for a long time—why is no one talking about it?
Here’s the source code between the doctype and the <title> for this site:
×26 <link rel="preload">
×36 external <link rel="stylesheet">
×91 external <script src>
This is not a nuanced point I’m making. We’re not debating the finer points of complicated web performance trade-offs. This is a very slow web site.
“This doesn’t look good” indeed.
@zachleat the original tweet in my post wasn’t fully that, but it shouldn’t be too much work to pull that info out and start putting it on a leaderboard
@zachleat actually, I’ve misremembered. It was just the homepages of frameworks. I’ll have a play this weekend and see if I can start putting something together.
Do you have any Saas services in mind? I remember Vercel having it on their site before
@hasanhaja I’d love to see a hosting leaderboard, yeah!
@zachleat maybe the website REACTs a litle bit too slow?
@zachleat I do love the idea of holding marketing teams to the same standards as their product… at a minimum (a marketing site *should* be way, way easier to make fast and accessible)
It's not just their marketing site. All of Vercel's claims fall apart as soon as you open Chrome and turn network throttle to "Fast 3G": https://x.com/dmitriid/status/1735338533303259571
@zachleat I have found that the real indictment is the showcase section of these sites. The speed of the things they choose to crow about reflects the lived values of these orgs.
@slightlyoff $300M buys a lot of marketing momentum
@zachleat If only it bought speed.
@slightlyoff I used to think the go-to metaphor was trucks https://fediverse.zachleat.com/@zachleat/110181262326744689 but now I think I’m going to use US passenger rail—slow, ineffective, expensive…
@zachleat RSC SSG prerender edge blazing fast AI PPR shadcn
Do you want more buzzwords or should I stop there?
@zachleat What do you mean that's slow? We're doing what you said right, splitting everything into chunks! Look how cacheable it is! HTTP/2 says it's good! Why don't you like it Zach! Why! Why! Why!
@zachleat What kind of intrigues me is that on 4G throttling and up, it really does feel fast, which I wasn’t expecting given all the server calls. HTTP/2 packing the calls into a single connection, I guess?
On Good 3G throttling and below, though, it really is deeply painful.
@Meyerweb The digital divide! I do think that there is a larger point worth making, as Vercel heavily invests in LLM automation of front-ends—automating this level of quality will only exacerbate the divide
The other point I’d make is that Network throttling is only one piece. Testing on a slow device is another thing entirely.
@zachleat @Meyerweb I’ve been working on an article calling for an inflation rate for the FCC’s definition of broadband. This is a great example of how the web gets more bloated not just in raw transfer size, but in bad habits that hamper limited data plans at an outsized rate. The 25 megabit plan of 5 years ago is slower to use today.
@slightlyoff @zachleat More like treo.sh/itespeed/, amirite? HEEEYOOOOO
@zachleat it’s a lot on the initial load, yes, but what you have to understand is that once all that is executed, it’s still slow.
@theadmiral this broke me
@theadmiral @zachleat I was getting angry until the last few words. Glad I persevered.
@robjwood @theadmiral we all went on this roller coaster together
@zachleat post this on the bird site... I dare you
@ryantownsend you->
<- me
where’d the go
@zachleat "Script tags are my passion"?
I've been doing (aka raging, ranting etc.) it for a while now on Twitter.
The result is a few retweets, and complete silence (or blocks) from the people responsible for this.
Latest is Googlers congratulating each other on the merger of Angular with amazing extremely performant innovative internal Google library called Wiz that, among other things, powers Youtube.
@zachleat I would imagine there are some recommendations for number of bytes before <title> too. Considering that so much stuff in head affects parsing behavior? Should want to get to it quickly so UA can render it in tab/window etc.
@zachleat “Let's add some code to make it faster.”
@zachleat I’m impressed.
This is art.
@zachleat Someone’s not getting a Vercel Christmas card this year.
@khalidabuhakmeh if they make next.js fast that would be a nice christmas present
@zachleat Dude… Santa’s not real.
@khalidabuhakmeh first of all—what
@zachleat No Christmas Miracles are happening this year!
@zachleat Here is my performance rates :D
Inside Devtools: 91%
Inside Netlify: calc(91% - howManyYearOldAmI - currentYearInSeconds - etc.)
@zachleat I'm planning on making a pitch to my tech lead to move off Ne*t.js so I'm actually gonna add this screenshot to the doc, thanks.
@mokobien good luck!
@zachleat yikes
I'd guess they* just don't care, I mean, it would track with how marketed towards the 'ai' hype it is
*Those that use vercel
@Ember I don’t think we need to critique the people—but the code is worth looking at!
@zachleat can we all already agree that react fucking sucks ? absurd "patterns", complicated API to work with and bad performance by default
I don't understand how react become the industry standard of javascript UI libraries when vue, svelte are much more ergonomic
@nrk9819 I won’t argue with you on these points
@zachleat We track a bunch of tools similar to ours (or at least those that haven't blocked our test agents), and yeah, there are some surprising results.
@tammy Wow, blocking your test agents! Wild!
@zachleat I guess this explains why they need over 100 pops, probably easier to battle the speed of light than making this performant
@zachleat This post is super important! I don’t know if everyone realizes just how much you genuinely care about pushing our collective baseline back to something reasonable. It’s too easy to dismiss this critique as competitive, but I have literally never seen you speak out about anything trivial.
Case in point, you could have posted about the massive performance regression I pushed to the @astro website last week, but you gave us a heads up instead.
@zachleat I'd love to see Vercel's answer, but keep in mind that a bare NextJS website is *fast*. You hit 90/100 without any optimization.
Also, Vercel employs the main dev of Svelte. Vercel's not only backing React/Next.
I don't care about Vercel.com's (landing page) google score (yes, keep in mind it's google score, and it doesn't affect SEO that much) as long as their tools and services are good.
@maeool a bare website with a 90 perf score is *not* fast
@zachleat Depends of your users. On a 56k modem no. In France, yes of course.
@maeool might want to re-read this one from 2012: https://blog.chriszacharias.com/page-weight-matters some important lessons in there!
@zachleat How is it relevant for vercel.com, the home page, or for my use case, France ?
It's also 15 years late.
Of course I understand the point though. Of course if your target is the entire world, and people with a slow connection are a significant part of your audience or potential audience, build your website for them.