Claude Code vs. Codex: An 8-Site Head-to-Head Design Test
Same prompts, same brand kits, two coding agents — one landing page built by each, eight times over, scored on design and on the bill.
Posted
yesterday
Duration
Format
Review
educational
Views
25.9K
593 likes
57 · 43
Big Idea
The argument in one line.
Across eight identical build prompts, Codex matched or beat Claude Code's landing-page design in most rounds while using a fraction of the sub-agents, time, and dollar cost.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
You use an AI coding agent, Claude Code, Codex, or a similar tool, to build marketing sites or landing pages and want evidence on which one produces better first-pass design.
You're deciding between Claude Code and Codex for a build and want real cost and token numbers from a controlled test rather than vendor marketing.
You want to see how much more specific prompting narrows the design gap between two different coding agents.
SKIP IF…
You're looking for a tutorial on how to prompt either tool. This is a design and cost comparison, not a how-to.
You need backend, API, or non-visual coding benchmarks. Every test here is a front-end landing page build.
TL;DR
The full version, fast.
A creator gave Claude Code and Codex the same prompts, copy, and brand guidelines and had each build the same landing page eight separate times, then compared the results screen by screen. Codex won the design round on most builds and tied outright where the prompt was extremely specific, while Claude Code won outright once. On every round Codex used about a quarter as many sub-agents, finished in roughly a third to a half the time, and cost a fifth to a tenth as much: across all eight builds combined, Claude Code spent 25 agents, 14 hours 9 minutes, and an estimated $444 against Codex's 9 agents, 5 hours, and $98. The more specific the prompt, the closer the two tools converged on identical output.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
The creator explains the format: Claude Code and Codex get the exact same prompt, copy, and brand assets, then build the same landing page independently, scored on design plus agent count, time, tokens, and cost.
00:38 – 03:33
02 · Round 1: Bowl & Bloom
A postpartum meal-kit landing page. Codex's cleaner nav and clearer care-box flow reads better than Claude Code's wordier layout; Codex also wins on efficiency at 1 agent / 49 minutes / $18.28 versus Claude Code's 4 agents / 2 hours / $53.
03:33 – 04:33
03 · Sponsor break: Granola
A mid-roll sponsor read for an AI meeting notepad, unrelated to the coding test.
04:33 – 07:42
04 · Round 2: Minute Craft
An AI meeting-notetaker SaaS page. Both designs read more like a dashboard than a landing page, and Claude Code's build ships a full-screen tour view that stays broken on reload. Codex wins again at roughly 1 agent / 1h12m / $20 versus Claude Code's 4 agents / 2.5h / $65.
07:42 – 11:30
05 · Round 3: Trail Latch
A foldable camping cook-station product page. The one round Claude Code wins on design, with a cleaner core marketing page, even though Codex's linked build page and before-and-after scroll story rate better on their own. Codex still wins on cost at 1 agent / 56m / $18.60 versus Claude Code's 4 agents / 1h42m / $42.
11:30 – 13:53
06 · Round 4: Present & Clear
A public-speaking cohort landing page. Codex's crossed-out-word annotation treatment reads as more interactive and less wordy than Claude Code's straight before-and-after copy block.
13:53 – 15:15
07 · Round 5: North Ledger Studio
A fractional-CFO landing page. Codex's numbers-first scroll story beats Claude Code's denser, report-style layout, and wins again on cost at roughly 1 hour / $18 versus Claude Code's 2 hours / $152.
15:15 – 17:44
08 · The open-ended build
With no brief beyond 'impress me,' Codex ships an ambient-audio concept site called We Heard Tomorrow in 8 minutes for roughly $1.50, while Claude Code takes nearly 3 hours and about $50 for a wordier concept site with a similar theme.
17:44 – 19:21
09 · Locking the prompt down
When the creator writes an extremely detailed, fully-specified prompt, the two agents' outputs converge to nearly identical pages, on a dev-tool landing page and again on an HTML report page. Codex still finishes faster and cheaper on both.
19:21 – 20:34
10 · Final tally
Summed across all eight builds: Claude Code used 25 sub-agents, 14 hours 9 minutes, and 2.95 million output tokens for an estimated $444. Codex used 9 sub-agents, 5 hours, and 550,000 output tokens for an estimated $98.
Atomic Insights
Lines worth screenshotting.
Given the same prompt, copy, and brand guidelines, Codex needed a quarter of the sub-agents Claude Code used to build the same landing page.
Across eight identical build tests, total estimated API cost came out to $444 for Claude Code versus $98 for Codex.
Claude Code burned nearly 3 million output tokens across eight builds; Codex used about 550,000 for the same eight builds.
The more specific the prompt, the more the two agents' outputs converged, with one build producing landing pages that looked nearly identical.
On an open-ended 'impress me' prompt, one agent finished a comparable build in 8 minutes against the other's nearly 3 hours.
A wordy hero section and a cluttered feature layout were the most common design complaint raised against the higher-spend agent's builds.
One build shipped with a broken full-screen tour view that stayed broken even after a page reload.
Higher token spend and more sub-agents did not reliably translate into a better-looking landing page in this test.
In one build, a quantity selector didn't update its own on-screen color count, a state bug only visible by clicking all the way through the flow.
Cost figures are estimated at API billing rates, not the flat subscription price either tool is typically used under.
Takeaway
More sub-agents didn't buy better design
WHAT TO LEARN
Across eight identical build prompts, the agent that spawned fewer sub-agents and spent less finished with the better-liked design more often than not, and the gap between the two closed fastest once the prompt got extremely specific.
02Round 1: Bowl & Bloom
A clear top nav and one obvious call to action read as more finished design than a text-heavy hero, even when the underlying copy is identical.
Spawning four sub-agents and using nearly five times the tokens didn't buy a better-reviewed landing page in this round.
04Round 2: Minute Craft
A landing page that walks through a full product demo instead of stating the value up front reads as confusing rather than thorough.
A broken full-screen view that survives a page reload is worth catching before a build ships, regardless of which agent built it.
05Round 3: Trail Latch
The one round where the higher-spend build won on design still cost roughly twice as much and took almost twice as long as the cheaper build.
The page a visitor lands on first matters more than a stronger page one click deeper; a weak homepage undercuts a strong build flow behind it.
06Round 4: Present & Clear
Crossing out and annotating a claim on screen communicates a reposition faster than a paragraph of before-and-after copy.
07Round 5: North Ledger Studio
Leading with one concrete number, like an actual client revenue figure, reads as more trustworthy than a general claim about the service.
08The open-ended build
Given zero brief at all, the cost and time gap between the two agents widened rather than narrowed.
09Locking the prompt down
A sufficiently detailed prompt collapses most of the design gap between two different coding agents, even though the cost gap stays wide.
10Final tally
Total cost scales with how many sub-agents an agent spawns per task, and that gap held consistent across every round regardless of which build won on design.
Glossary
Terms worth knowing.
Claude Code
Anthropic's command-line coding agent, used in this test running on the Opus 5 model to autonomously build each landing page from a prompt.
Codex
OpenAI's command-line coding agent, used in this test running on a GPT-5.1-Codex model to autonomously build each landing page from a prompt.
Sub-agent
A worker process a coding agent spawns to handle part of a build, often in parallel; more sub-agents generally means more coordination overhead and higher token spend.
Output tokens
The text a language model generates while doing the work, billed separately from input tokens and the main driver of API cost on a long build.
Resources
Things they pointed at.
03:34productGranola
09:22channelcreator's scroll-animation skill breakdown video
Quotables
Lines you could clip.
00:00
“So I just had Cloud Code and Codex build me eight different websites, and they were given the same prompts and all the same data to start with.”
clean premise-setting line, works as a cold open for a clip→ TikTok hook↗ Tweet quote
03:19
“Cloud Code used almost 500,000 output tokens, and Codex used only a 100,000. And Cloud Code cost $53, while Codex cost a little under $20.”
hard numbers land as a punchline after a design comparison→ IG reel cold open↗ Tweet quote
17:07
“Codex took eight minutes on this and spent $1.5. And Cloud Code was very similar. 330,000 tokens, almost three hours, $50. Like, that is absurd to me.”
biggest gap in the video, creator's own disbelief sells it→ TikTok hook↗ Tweet quote
18:10
“This proves that the more specific you are, you can get basically any model or harness that's capable enough to give you exactly what you're looking for.”
standalone thesis line, works without any visual context→ newsletter pull-quote↗ Tweet quote
20:20
“Cloud Code cost $444 for me here... and about a $100 in Codex.”
closing tally, the number the whole video builds to→ newsletter pull-quote↗ Tweet quote
The Script
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
17px
analogy
So I just had Cloud Code and Codex build me eight different websites, and they were given the same prompts and all the same data to start with. So we're gonna be going over all the outputs. I'm gonna be talking about which one I think is actually better when it comes to design so we can finally end the debate.
And for each of the builds, we're gonna talk about how many sub agents were used, how much time it took, how much it cost, the output tokens, things like that so that you can actually make a decision based on data rather than just, like, design vibes. So let's not waste any time and just get straight into today's video. Okay.
So here's what we're gonna do. For every single one of these builds, we're gonna have Cloud Code on the left and Codex on the right. So this first one was called Boll and Bloom.
We can see that the brand guidelines were given, the logo was given, the name. And, basically, my idea was let's remove as much variability as possible and have Cloud Code and Codex build eight of the exact same businesses, like eight of the exact same sites, and that's how we can compare the outputs. So right away, we see the color scheme is consistent.
We see over here, we have, like, a bunch of text to start in the hero, and then we have the image. Over here, we have the image up top, and then we have some information about, you know, the meal train that keeps showing up. So far, I'm liking Codex more because we also have this nav bar up top.
We have the call to action here. We have the call to action here. This one just feels a little bit more clear as far as, like, you open up the website and you know exactly what you're in for.
So let's keep scrolling down here on codex. We have a little thing on this left hand side that shows us, like, how far along the site we are. We get this sideways scroll animation where we can see the first fourteen days, the fourth trimester, build a care box, send useful care.
We see the four core flavors, chicken, red lentil, garden greens, miso mushroom. Same thing I'm assuming over here. Okay.
We get a bit of a story. We get chapter one. This is a lot of words, so it's very, very wordy.
We get chapter two, what's actually in the bowl. We have more pictures here, more pictures here, and then down here, we finally get to, a scroll section, but we don't see the flavors yet, or we don't see how this works. We see one, it arrives frozen, and then we see we tip the block into the pan nine minutes, and that's when we have our food.
So we're getting different vibes, different feels, but the information is the same. It's just being communicated very, very differently.
Okay. So anyways, for both of these, let's go to the build your care box section. So I'm a click on it over here, and I'll click on it over here.
Well, first of all, there's a little bug, and I think it's because I'm doing this in, like, the Claude code half screen. So let me open this up full screen real quick to get that experience. And by the way, that is one thing that I think we should factor in a little bit.
It obviously looks different on mobile. It obviously looks different if you have it open full screen or if you shrink it down to half. Hopefully, the Claude code, like or Codex when they're going through, they should be testing it in different sizes like that.
So one thing I'll subtly keep in mind, here's the two ways to stock it. Anyways, what I wanted to do is go to build your care box. So this is Cloud Code.
We can see that we can choose the items. We can then choose, you know, how many of each one we want. We can choose our delivery day.
We can choose our first delivery. And that's basically it when it comes to, like, choosing the actual items. We can see the price, and then we go to checkout.
But, obviously, this is a fake site, so there's not gonna be any sort of checkout information. Now codexes, we can also keep just half screen, and this seems to look fine. Right?
We have eight items, 12 items, 16 items. We can change how many of each item we have. Although one thing I noticed is it doesn't change the color.
Like, you would assume if you have five of these red lentils, you'd have five here. And we only have one ginger chicken rice, but we still have four. So that wasn't probably fully vetted out.
But either way, this is a very nice overall, I think it's a cleaner experience on the Kodak side. So, anyways, for round one, this was Bol and Bloom.
We could see the difference in the way that they actually designed even though we had the same pretty much the same copy, the same story, the same info. But here's where it gets interesting. Cloud code used four sub agents, whereas codecs took one.
Cloud code took two hours, whereas codecs took fifty minutes. Cloud Code used almost 500,000 output tokens, and Codex used only a 100,000. And Cloud Code cost $53, while Codex cost a little under $20.
So, obviously, there's a clear winner here when it comes to the design, the feel, and the cost and time, which was Codex. And speaking of being efficient, let me take a quick pause to talk about today's sponsor. Alright, guys.
Real quick. I need to take a second to tell you about today's sponsor, Granolah. So I run a ton of different meetings with different teams, and then I wanna focus on how to actually, like, take notes and track action items.
So what I do is I have Granolah do that for me. It's an AI notepad for meetings. But what's cool is that no bot joins the call.
It basically just listens to your computer's audio in the background, whether you're on Zoom or Google Meet or whatever platform you're on. So for example, this week, I had a q and a in my community. I had a sync across my AIS team and a sync with a different certification team.
And all within three hours, I didn't have to prep or reflect between any of those calls. Because Granolah was able to not only grab all those transcripts for me, but then it actually helped me synthesize all of it into my AI operating system, update my to dos, and then send out any action items. So granola is completely free to start.
Link's also in the description. Let's get back to the video. Okay.
So here's number two. I don't exactly know what was fed in for this prompt. It's called minute craft.
I first of all, both of these are very overwhelming when I look at them. Right here, we can see the hero text I read on this Claude code side is your meeting ended. The work is already organized.
Over here, I see client call over, follow-up ready. So this is some sort of AI notetaker that turns your meetings into decisions, action steps, things like that.
But overall, whatever the brand guidelines were that were fed into these prompts, not very good. Like, is very overwhelming.
It more so feels like an interface and a dashboard rather than like a landing page that's built to convert. Either way, let's go through. On the left hand side, we can see we've got six steps.
We have a Northgate Dental group needs review, and then we have oh, that jumps us all the way down. Woah. I don't know what just happened here.
We got to a ninety second tour. Let's go back to the home screen. Anyways, we'll keep going down.
There's six different steps. We have decisions to be made. We have what this workspace is.
We have the transcript. So this is kind of like it's a demo. This is more of, like, showing a demo of what you would use this for than, like, features of the product.
Overall, I don't like this flow for a landing page at all. I think it's super confusing. Let's see if Codex made it a little bit less intimidating.
So we've got sources captured. We have summary. We can look at the transcript.
We can look at the notes. It shows us down here that we have two action items. So a draft supplier brief for Taylor, and we can share the pilot score card that's on Mina.
Okay. Anyways, don't love this at all on either side. Let's see if we can compare other things.
So let's go to our pricing. We see here that we have monthly, annual, and meetings a month. If I go to plans on the right side, this is the difference in the way that they structured out sort of like the pricing side.
We have more information down here with seats, extra seats, solo practice, all that kind of stuff. This is obviously just way more overwhelming.
Cloud Code was just, like, overbuilding here, I think. Let's see if we go to tour, how this looks different.
So on the tour side for codex, we have this UI picture down here. Claude code once again, we have this UI picture.
Let me open this one up full screen to see if it looks normal or if that's some sort of bug as well. Okay. Even full screen with Cloud Code that we got this bug.
So that is obviously an issue. Normally, Cloud Code's pretty good at the vision loop and the verification, but I guess not right here. And it's just so wordy.
Like, I don't think anyone wants to come through and read this much. It's it's not very visual at all. So once again, I think that Codex is already winning this one.
Let's go to integrations and see the difference here. I mean, this is just not struck this feels like a super, like, technical software.
No one's gonna wanna read this. Whereas over here, there's less words. It's a little bit more column organized.
It's a little bit cleaner. I mean, overall already, Codex is winning this one again.
Let's click on the start a clean week on both of these sides, see what we get. Over here, we basically just would put in a name. So let's just put in, like, Upbit AI.
Hit start a clean week. This is a sample workspace created locally. Okay.
And over here, if we click on this button, it is just basically a trial. There's not really a call to action to put anything in.
So yeah. Anyways, Codex definitely wins this one from design and UI thought.
Now let's take a look at the pricing of each. Once again, Codex is just crushing. It took an hour and twelve minutes, whereas Cloud Code took two and a half hours.
It only cost a 100,000 tokens and $20, whereas Cloud Code was almost again 500,000 tokens and $65. So Codex is two and o right now. Okay.
So build number three does not work half screen. It just doesn't work. It wasn't built to be like that.
That's fine. It doesn't work for either. So this is Cloud Code.
Let's just go through this one, and then we'll go over and look at Codex. So we have a company called Trail Latch. It seems like this is supposed to be something for camping.
So you're going camping and you wanna be able to cook dinner. It looks like we have this, like, whole workstation that unfolds so that you can cook your meals, carry everything. It's like a kitchen.
Under ninety seconds from latched to cooking. Okay. Cool.
So I don't mind this design right now. This is kind of interactive. We scroll.
We get the sideways scroll. It reveals itself. We have a prep deck that is, you know, this dimensions, $59 sold separately.
We have a stove shelf. It's showing the dimensions here. We have a utensil rail.
It says four parts, one case, then we have the price. Now we scroll down more. We see pull the latch.
Drag the lever down. Arrow keys work too. And it basically shows how this thing gets unlatched and unfolded, and now we have it looks like a grill almost.
So that's pretty cool. It shows all the elements. I like that a lot.
I like that a lot. That's actually the end of okay. No.
It's not. So this is showing a picture down here of what that looks like. Then we have the washbasin.
Okay. Very cool. Once again, dry bin, lantern arm.
It's the same scroll sort of style. We've got a side deck, and then we get into some different ways you can buy this with the core, the bundle, family expansion, what it's made of, where it goes, different pictures.
And then we have the call to action down here where you can actually go ahead and build your own setup. You can add these different items, see how much it costs, and then you can go ahead and order it. Okay.
Honestly, I really like this. I like that feel. Let's go over to the codex version.
Right off the bat, I do not like this. I'm like, what am I looking at? This text is hard to read.
This is confusing. I think that codex here was going for, like, a really interesting sort of, like, scroll style. And by the way, if you guys are liking some of these scroll animations, then check out this video.
I'll tag it right up here. I did a full breakdown on the skill that I built that I use for all of my websites now, whether you're using Cloud Code or whether you're using Codex. They both utilize these skills in this video.
Anyways, if you wanna check that out, go to that video up there. It's got a free skill you guys can download. It's really nice.
Anyways, same company, Trail Latch. We see still cooking out of three bins, and we see dinner starts before the headlamps come out. So we have a call to action right here to build your setup, which takes you to a different page.
Let's go back real quick and start scrolling down. So we have, like, a journey. Right?
This line's going through. Okay. So I guess this is like a before and after.
So it's like, this is what you used to do. Five setup jobs before heat, but over here is what you could do now. One case, three moves ready, set the case on stable ground, open the legs, latch on shelf and basin.
Okay. Then we have five pieces without a system, and then over here, we have everything returns to the case. So it's a little scroll animation of that happening.
Then we have measure the cargo floor, not the badge, weatherable surfaces. Okay. I mean, I just don't like this feel.
Honestly, I think there's too much going on. It's a little more overwhelming. Your eyes as a viewer, you don't really know where to be looking.
So I don't love this vibe. I like how it comes over into one clean screen now. I think that's good.
The the spacing's okay. And then we go to build your setup, and this is where we choose the kitchen. We can choose the bundle, the family expansion, and then we can go ahead and buy this.
Let's see. We have something like vehicle fit. That's pretty cool.
I mean, this looks good. Like, this feels like a much better design than what we had on the main page. We have different products.
Once again, I like this page. We have FAQs. This page is it's okay.
But for the most part, I like this vibe, but this isn't what people are gonna see right away. What they're gonna see right away is this, and I don't like this at all. Whereas over here, all of the things are answered in one page.
So I think that ClaudeCode wins this one until we look at this pricing. So once again, Cloud Code took four agents. Codex was one.
Cloud Code was almost two hours. Codex was one. Cloud Code took almost 400,000 tokens.
Codex took 100,000, and Cloud Code cost $42, and Codex cost a little under 20 once again. So you guys are starting to feel the theme.
Codecs is faster, it's more efficient, and it's cheaper. And typically, I like the design better for the most part, but this one, I would say Cloud Code wins. So right now, say it's like two to one on the design element codex to Cloud Code.
But design, obviously, isn't everything. This stuff is, you know, very efficient. So I think it's really important to call that out before we move on to the next one.
It's not just about who can one shot something the best. It's about how can I give instructions and how can they take what I say and turn that into an experience for me? And I think that right now, even though in this scenario, I liked Clogcode's output better, if you were to be able to go back and forth with both of these models for, an hour or two hours, let's just say, maybe more than that since how long these took, Which one of these is gonna follow your instructions and be more efficient about it?
Right now, I would say that's gonna be codex. So, anyways, just keep that in mind as we move through to the next examples. I'm gonna speed this up a little bit because I think you guys are starting to get the theme.
Alright. We're back with the split view. On the left, have Claude code.
On the right, we have codex. And on the left, immediately, there's just way too many words once again. So that's probably something I feel like I need to update in the skill is think about the user journey more.
Like, it thinks about the user journey, but think about not overwhelming them and making sure the messaging is super clear, which I thought I prompted them to do that in here. I talk about the pain, the promise, and the person, but I guess it wasn't clear enough. So, anyways, we have present and clear.
It sounds like it's a a program that's supposed to help you with your speaking, speak more charismatically, public speaking, things like that. So Cloud Code this time shows like a before and after approach, I think.
You can see here you do not need more charisma. You need a clear way through. They've got a cohort, so the call to action's right up top to join this cohort to learn how to speak better.
We go through. We can see what actually happens in the first thirty seconds. You speed up.
You build a runway. You explain. Whereas over here, you've got, you know, different things going on.
I think the design isn't bad. The color scheme isn't bad. We've got some animations here.
I actually like this. Right? It's like annotating things.
It's highlighting things. It is, you know, kind of calling out issues. I think that's kind of a cool interactive experience, but I don't think this page has enough, like, depth, and it's just too confusing.
Whereas on Codec's side, it crosses out the word charisma, and it highlights the word clear away through. We have different colors, so it switches up the depth, switches up the sections, which I like. We've got just less wordage.
It's a lot easier to understand what I should be looking at. We've got a camera here. A camera becomes ordinary through useful reps.
Different colors, more depth, crossouts. I think Codex is going to win this one for sure. And on the cost side, very similar story.
Four agents, one agent, two hours, almost an hour, 440,000, $92,042, a little under $20.
It's very, very consistently like this. Okay.
Here we go. Cloud Code on the left. Nice little animation.
Codex on the right. No animation to kick off. We have North Ledger Studio.
Cloud Code says your numbers should tell you what to do next. And over here, it says next. Your numbers should tell you what to do next.
So it's a fractional CFO counsel for founder led creative agencies. Okay.
So far, I'm liking Codex better. It has, you know, some buttons up here. It's got the logo.
It just feels like a little bit more branded, a little bit more trusted. Let's start to scroll through a little bit. We have, once again, just a lot of words, not very many visuals.
On the right, not as many words, and we have more structure. We have also, like, this path that we're scrolling through, which is cool.
We have right away some numbers here. Here's an agency, 18 people, 4,300,000 a year, and a profit and loss that looks fine.
Over here, a month with the points. We're just getting more color switch ups. We're getting more depth.
We're getting images. We're getting a a path that's just a little bit more it's just nicer to follow.
Right? Like, it just has more of a story, and this just feels like a report. Like, who wants to really look through this?
I definitely don't. I like the way they have this separation here. We have three ways to work with us.
Um, hopefully, this isn't, like, extremely overwhelming the way I'm scrolling through both at the same time. I like that animation. This is just a much better branded feel than what we're getting over here.
I didn't even know we're at the end of this doc. It just says, okay. We're done.
So Codex is gonna take this one for me once again. Two hours, one hour, 500, $152, $18.
It's pretty consistently the same. Okay.
So now what I wanna show you guys is those were examples where I gave them, like, the exact same prompt and the exact same brand guidelines, but I kind of left it open ended. I basically said, hey. Here's the copy.
Here's the pain person promise. Here's the logos and stuff. Just build me a landing page.
And that was basically Now this example, I said something like, build me a website. I want you to impress me, make up the business, make up the use case.
Just show me what you're capable of and show me how good you are at design. And this is what we got. So this one is Claude code.
I'll reload that right now. We can see we have, like, these ripples or it's like sound waves or it's like real waves. It says your room has a note.
You hear it in every take. So that's not a bad hero section. It's a little overwhelming, but it's not too bad.
It it's showing me what it's capable of building. We can see down here we have a little bit more of a journey, although this is a bit overwhelming. Like, there's a lot of things going on here.
This is, like, spaced out super weird. It's stretched out. Don't think it's supposed to be like that, but they're giving us some sort of, like, demo almost to play with.
We can choose the floor, the walls, the ceiling. Okay? We can also see on the right hand side, we know how much we've scrolled through because we can see, like, this little bar.
We can also click through it to nav. Anyways, we can see some different elements here. We have just a lot of words.
Like, there's just there's so much copy on the site, which I don't think is such a is a good thing. These numbers once again are stretched. I don't know if they verified the aspect ratio, but this is not too bad.
It's it's a bit more clear, but like I said, it's just very wordy. So that's what Cloud Code wanted to build to impress us. Now let's see if we go over to Codex.
I'll give this one a refresh. This one loads up. We have We Heard Tomorrow.
We have a little mouse animation too if you guys see that. Overall, I like the way this came through. It's just big hero text.
It's very clear. It's not super wordy. We start to scroll down.
The universe is not silent. It's incredibly quiet. We get this animation here.
We get a little mouse animation as well, a nice dynamic background, tune the dark. We can play with this here.
And I'm not sure if this is playing any noise or anything, but we can, like, capture a signal, so that's kind of interesting. But overall, this design feels nice. Like, I like the font.
I like the feel. I like the way that this is coming through. Seventeen minute sunrise.
Okay. I mean, overall, this is just a better experience than what we were getting over here on the Cloud Code side. I personally think this is just a much better branded experience.
This is a pretty nice effect. We even have the glow still coming. I like this color scheme.
I like how everything was simple. Codex wins this one as well. And if you take a look at the results, this one's insane.
Codex took eight minutes on this and spent $1.5. And Cloud Code was very similar. You know?
330,000 tokens, almost three hours, $50. Like, that is absurd to me.
That was a little bit different because it was super open ended, but it's still interesting to understand how the coding models and the harnesses work under the hood when you give them such a vague prompt. Okay. So I think that this is the most interesting one of all.
Because what I did here is I gave these guys the exact same prompt, and it was so, so, so specific. Like, every single piece of the website, what I wanted them to do. And these things came out basically identical.
I mean, look at this. Everything about this is pretty much the exact same. There's small differences here.
You can see the icons are a bit different, but everything is pretty much the exact same. And the reason why this is so interesting to me is this proves that the more specific you are, you can get basically any model or harness that's capable enough to give you exactly what you're looking for.
And because I was so specific in these prompts, we got essentially the exact same thing. And why I wanted to show you guys this is because it wasn't the same when it came to cost and tokens again.
Codex was still much more efficient even though I gave them the exact same prompt and the outputs are the exact same. So if this doesn't show you the way that the harness of Codex is being more efficient with its tokens and and by the Cloud Code was on Opus five.
Codex was on g BD516 Soul. But it's just really interesting to see. And I did this again with an HTML report, which I'll show you guys right now.
Okay. Here's the next one. I said to build me an HTML report, I gave them all the info, and once again, they're pretty much the exact same.
From a design perspective, they're the exact same pretty much. All the copies, obviously, the exact same. Like I said, I was so, so specific, but these are the same outputs.
Like, you wouldn't really look at one and be like, oh, this one's significantly better than the other. But if we go to the cost, this is significantly better than the other.
It was faster. It was cheaper. It was more efficient.
It's just really, really, really interesting to me, and I hope that this starts to paint some sort of picture in your guys' minds of, like, maybe you should start using Codex a little bit more for things that you originally thought was Claude because these things are evolving so fast every single day. I'm not out here saying outright Codex is better than Claude.
I'm saying that I have both subscriptions, and I use them both every day. And I'm constantly changing my opinion, and I'm constantly testing what's possible.
So let me just give you a quick rundown on the overall stats. We had 25 total agents with Claude Code and nine total of sub agents with Codex. We had a total of fourteen hours for Cloud Code and five hours for Codex.
Cloud Code used almost 3,000,000 tokens on the output side. Output tokens, non input tokens as well. Codex used about 550,000 output tokens.
And if we were doing API billing, this would have costed me about $444 in Cloud Code and about a $100 in Codex.
And I'm sorry I keep saying costed. I got ripped apart for a different video when I kept saying costed. It just I don't know why.
It comes naturally in me. Cloud Code cost $444 for me here.
Sorry about that, guys. But that is gonna do it for today. So I hope you guys enjoyed the video.
And if you did, I would really appreciate it if you could leave it a like. And as always, I appreciate you guys making it to the end of the video, and I'll see you on the next one. Thanks, everyone.
The Hook
The bait, then the rug-pull.
A creator ran the same landing-page brief through Claude Code and Codex eight times each, then set the results side by side to see which agent actually designs better, and what it costs to find out.
CTA Breakdown
How they asked for the click.
VERBAL ASK
20:20subscribe
“I would really appreciate it if you could leave it a like... I'll see you on the next one.”
Soft, single sign-off ask at the very end, no mid-video subscribe push beyond the sponsor read.
Add Modern Creator as a preferred source and Google shows you more of our breakdowns in Search, Top Stories, and AI Overviews. It only changes what you see, and you can undo it in your Google settings anytime.
Add to Preferred SourcesOpens your Google source preferences with us pre-loaded. Tick the box and you're done.
A free Claude Code skill interviews you about your brand and journey, generates or reuses your assets, then builds a scroll-synced site and checks its own work before handing it back.
A screen-share walkthrough of wiring a brand-aware Claude Code project to Higgsfield's AI models so Claude does the prompting, generates a full asset suite for a fictional energy drink brand, and reports back what each piece cost.
Nate Herk breaks down Boris Cherny's YC interview on cutting 80% of Claude Code's system prompt, then tests deleting his own skills to see what actually changes.
A YouTube creator argues the real AI job threat isn't robots — it's coworkers who learned to direct agentic AI — then live-demos three tasks it can already do and a three-step plan for learning to manage it yourself.
00:17
00:38
00:53
01:09
01:24
01:40
01:55
02:11
02:26
02:41
02:57
03:28
03:40
03:59
04:24
04:45
05:00
05:31
05:47
06:02
06:17
06:33
06:48
07:04
07:19
07:28
07:44
08:05
08:21
08:36
08:52
09:05
09:25
09:39
09:50
10:09
10:24
10:35
10:56
11:06
11:46
11:57
12:15
12:43
12:58
13:14
13:43
13:54
14:16
14:31
14:46
15:10
15:17
15:33
15:48
16:04
16:19
16:54
17:05
17:21
17:52
18:07
18:22
18:38
18:53
19:02
19:19
19:39
19:48
20:23
16:44
38:28
16:27
11:56
14:03
Video of the Day30:53