Modern Creator
Pat Simmons · YouTube

I Made Opus 4.8 and Fable 5 Build the Same App (RAW RESULTS)

Three identical one-shot prompts. Two models. The gap was not close.

Posted
today
Duration
Format
Demo
educational
Views
89.5K
3.1K likes
Big Idea

The argument in one line.

Fable 5 does not just produce better output than Opus 4.8 -- it self-corrects on visual design, orchestrates parallel agents, and finishes 15 to 20 minutes faster, making the gap feel generational rather than incremental.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…
  • You are deciding whether to move from Opus 4.8 to Fable 5 for agentic coding work and want raw, unedited session evidence rather than benchmark scores.
  • You use Cursor with Claude and want to see how the two models behave differently as agent orchestrators on long multi-file builds.
  • You are building something visual in a single session and care about how well the model prompts downstream image generators.
  • You are on a Claude Max plan and want a real-world cost-per-build estimate before usage-based pricing activates.
SKIP IF…
  • You need controlled benchmarks -- this is one developer session, not a systematic study.
  • You are evaluating models for writing, summarization, or analysis -- every test here is agentic coding.
TL;DR

The full version, fast.

Same prompt, three builds, two models side by side in Cursor: Fable 5 consistently outpaces Opus 4.8 on both speed and quality. For the e-commerce build it produced a cleaner storefront with better AI-generated product photography, finishing 15 minutes faster. For the 3D art museum it delivered a fully navigable gallery with nearly 800 Wikipedia paintings, while Opus 4.8 gallery click-to-enter broke due to a canvas event conflict. For the Age of Empires clone Fable 5 shipped a playable 3D RTS with realistic textures; Opus 4.8 produced broken blobs. Fable 5 costs roughly 37% more per build but uses fewer output tokens, and its self-directed design corrections and autonomous agent spawning set it apart beyond raw output quality.

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.

Create a free account →
Chapters

Where the time goes.

00:0000:23

01 · Cold open + premise

Three builds, same rules, results were not even close.

00:2402:47

02 · Build 01: E-commerce setup

Slow Burn candle store prompt, usage tracker intro, both models suggest same three brand directions.

02:4704:39

03 · Opus 4.8 e-commerce review

Readable but imperfect: awkward hero placement, clunky nav, overly granular filters.

04:3907:00

04 · Fable 5 e-commerce review

Cleaner Shopify aesthetic, better GPT Image 2 prompting, ritual-based filters, polished footer.

07:0007:59

05 · Build 01 cost breakdown

Opus $21.41 vs Fable 5 $36.84; Fable 5 15 min faster and more token-efficient.

08:0110:27

06 · Build 02: 3D art museum setup

Zoomable art-history timeline, Wikipedia API, Neon DB, realistic Three.js galleries.

10:2713:05

07 · Live build narration

Fable 5 self-corrects lighting, spawns parallel agents, handles Wikipedia edge cases.

13:0514:07

08 · Opus 4.8 museum review

Atlas of Art -- color-coded timeline works, but gallery click-to-enter is broken.

14:0716:52

09 · Fable 5 museum review

Star-map timeline, 767 paintings, walkable Degas gallery, GSAP animations.

16:5217:22

10 · Build 02 cost breakdown

Opus $46 vs Fable 5 $64; Fable 5 37% more expensive but 36% fewer output tokens.

17:2318:35

11 · Build 03: Age of Empires setup

Full 3D RTS in the browser, Three.js, town center, enemies, buildings.

18:3519:14

12 · Opus 4.8 game review

Broken blobs, no camera movement, non-functional controls.

19:1520:48

13 · Fable 5 game review

Empires of Dawn -- realistic terrain, navigable map, buildable structures, armored enemies.

20:4921:11

14 · Closing

Quality bar is now set; Opus 4.8 cannot match it. Anthropic planned this.

Atomic Insights

Lines worth screenshotting.

  • Fable 5 finished all three builds 15 to 20 minutes faster than Opus 4.8 on identical prompts.
  • Fewer output tokens does not mean less code -- Fable 5 was more token-efficient and still produced more working functionality.
  • A model that proactively corrects its own visual design decisions without being asked is qualitatively different from one that only executes instructions.
  • The 3D art museum ingested 767 paintings from Wikipedia into a Neon database in roughly 30 minutes from a single one-shot prompt.
  • Fable 5 costs approximately 37% more per build than Opus 4.8 at equivalent complexity, due to higher per-token pricing despite lower token count.
  • Opus 4.8 and Fable 5 independently suggested the same three brand direction names for the candle store.
  • An Age of Empires-quality browser RTS with a navigable 3D world and working unit controls was produced in one shot in 30 minutes.
  • The $64 cost for a 3D museum with nearly 1,000 Wikipedia paintings is roughly equivalent to one month of a Shopify subscription.
  • Two complex builds burned through 40% of a Claude Max 20x subscription in 30 minutes.
  • Fable 5 autonomous agent spawning and self-queuing during the museum build represents a step-change in long agentic sessions.
Takeaway

What the Fable 5 gap means for builders.

WHAT TO LEARN

A faster, more token-efficient model that corrects its own design decisions mid-build changes what you can expect from a single-session prompt.

  • Speed compounds: a 15-minute difference per build adds up fast when you iterate on multiple projects in a day.
  • Token efficiency is not the same as output quality -- Fable 5 used fewer output tokens on every build while producing more working code.
  • A model that proactively flags and fixes its own visual mistakes is operating as a collaborator, not just an executor.
  • One-shot agent orchestration -- spawning parallel sub-agents and self-coordinating -- is now a real capability, not a demo trick.
  • Usage-based pricing makes cost-per-build a first-class decision: $21 vs. $37 for an e-commerce site, $46 vs. $64 for a database-backed 3D app.
  • Image prompting quality is now an emergent model capability -- neither model was given instructions on how to prompt GPT Image 2, yet outputs differed dramatically.
  • The same high-level prompt resulted in one model producing a broken game and the other a playable 3D RTS -- the prompt was not the variable.
  • Subscription caps become a real operational constraint: two 30-minute builds can consume 40% of a Claude Max 20x plan.
Glossary

Terms worth knowing.

One-shot
A build completed from a single prompt with no revisions -- the model must plan, implement, debug, and deploy entirely on its own.
Dynamic workflows
A Claude feature that splits a large task across a swarm of parallel agents; tested in a prior video and found underwhelming for the same e-commerce build.
Neon DB
A serverless Postgres database used in the museum build to store artist and painting records extracted from Wikipedia.
GSAP
GreenSock Animation Platform -- a JavaScript animation library used for smooth card transitions in the museum timeline.
GPT Image 2
The image generation model used by both Claude models to produce the 30 product images for the Slow Burn candle store.
Claude Max 20x
An Anthropic subscription tier with 20x the usage of the standard plan; both sessions ran on a single Max account.
Usage-based pricing
Anthropic pricing where charges are based on token consumption rather than a flat subscription rate -- Fable 5 transitions to this within weeks of recording.
Three.js
A JavaScript library for WebGL-based 3D rendering in the browser, used for both the art museum galleries and the Age of Empires clone.
Resources

Things they pointed at.

00:40productClaude dynamic workflows
08:40toolNeon DB
05:08toolGPT Image 2
Quotables

Lines you could clip.

00:20
The results were, I'll be honest, not even close.
tight one-liner, no setup neededTikTok hook↗ Tweet quote
10:56
The pitch black floor in the museum screenshot bothers me. Let me look at the floor and lighting settings. It is really quite opinionated -- the first time I've ever seen something like that.
concrete differentiator moment -- model shows taste, not just executionIG reel cold open↗ Tweet quote
20:40
We will now be getting used to a certain level of quality that just cannot be matched by Opus 4.8, which is surely Anthropic's plan all along.
punchy editorial closer with mild cynicismnewsletter 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.

00:00So I just gave the exact same prompt to Opus four eight and Claude's brand new flagship model Fable five across three builds, a fully functioning ecommerce store, an entire three d world of famous art pieces, and finally, an Age of Empires clone. Same rules for both models, a single one shot output, no revisions, deployed live in the Internet.
00:16This is Opus four eight versus Fable five head to head, and the results were, I'll be honest, not even close.
00:24First up, we are going to build out a full ecommerce site. So I actually ran this exact build in a recent video testing Claude's dynamic workflows, which is their new feature where Claude splits a big job across a swarm of parallel agents, and I was really underwhelmed with the performance. Even with the ultra code dynamic workflows that spawned like 30 plus agents, it got a store up, but then it missed a lot of the UI stuff.
00:44Like every product looked identical. There was no labels. There wasn't even a hero image on the landing page.
00:49So we're gonna run it again. We're gonna hand the same prompt to Fable five on the left and Opus four eight on the right, and we're gonna give Opus a chance to redeem itself. I did also revise the prompt a little to give them more of a chance.
00:59So calling out the product labeling this time and other stuff in the prompt sort of avoid some of these egregious mistakes that I made last time. And this one build is gonna test out a lot at once. So we have landing page and ecommerce design.
01:09We have the database underneath. We have the entire architecture of all of this. So this could actually be useful, especially if you're, I don't know, moving away from platform like Shopify or building out features or trying to build out a whole ecom platform.
01:19So we'll see how both of these do. So here is the prompt. Build me a complete e commerce store for a fictional brand called Slow Burn, a small batch candle company, production quality mimicking a real Shopify store, 30 distinct products, every product image must be visually distinct, all text must be readable, no low contrast text.
01:34Any of those pitfalls that I fell into last time, hopefully, will be missed. And I do have a check-in here even though it's technically not a one shot, but I just wanna see the different design directions that Fable five goes to versus Opus four eight. But that's the only tweaking we're gonna do.
01:46There's no there's no other revisions. So alright. It's already giving me quiet.
01:50Can't pronounce that. Apoth apothecary? Dark ritual modern wellness.
01:53It's already giving me ideas. Whatever. Okay.
01:55So I guess, yeah, fun. I bet Opus four eight is gonna do the same thing. By the way too, I built this usage tracker so we can see exactly the input tokens, output tokens, and the estimated costs if it were to be usage based pricing, which Fable five will go to in a couple weeks.
02:07So you can see the difference in costs there, and I'll also let you know how long each of these builds take. So we'll have it across the ecommerce store, three d army examine, age of empires. This is funny too.
02:15Opus four eight give the exact same options. Warm apothecary, darkritual, editorialcom as Fable five.
02:21Fable five is on the image generation step waiting for an agent to generate all 30 of these images. Claude is still moving through the rest of this build. So Fable five is faster than Opus four eight, dare I say.
02:33And Slowburn is live both from Fable five and Opus four eight. So we're gonna look at Opus four eight's output first. A little link here for opus four eight.
02:42And here is slow burn, rituals for people who work too much. Okay. Not bad.
02:47It's a little bit better than the dynamic workflows example. At least the text is readable. However, I see some little things here and they're already like small batch candle code.
02:56It just doesn't read well. The candle placement really should be placed on the right, not the left because it's hard to read over the text. That's something that, I don't know, I feel like it's relatively simple for the models to notice.
03:07And then we kinda had this weird CTA here. I don't love that button. But let's keep scrolling.
03:12Okay. It's the slow burn idea. You don't need another productivity app, yada yada.
03:15Okay. So it breaks it up by categories. Alright.
03:17Nice little nice little animation. Nothing real new there. Okay.
03:21The actual image generations, they look better. We actually have labels on these candles now, not like last time where there was no labels. And they're yeah.
03:30They're big. We got a nice little studio type setting. Nice little animation, little hover animation on these.
03:35We've got the catalog here. We'll click that in a second. Okay.
03:37And then we have some, like, you know, classic proof points. I don't love these these section colors and then nice lit clock out on a Tuesday and oh, clock out is then candle.
03:47Got it. And actually closed my laptop first time in a month. Attired customer.
03:50Nice. Okay. Nice.
03:51Nice. Little touch. And then we have the footer here.
03:52Alright. Yeah. Not bad.
03:53Let's check out the categories first. Okay. A little all over the place.
03:58Why would you think that this is a proper filter? I mean, are just so many. But okay.
04:04We can filter by alright. That's kind of weird. Unclick each of these.
04:08Blood orange did a good job of categorizing these candles. And then we have these categories based on I guess your use case, morning ritual, deep work, unwind, sleep.
04:17That's not bad. It's not bad. The featured tab here.
04:20What else? Okay. Let's just click into one of these candles.
04:23Alright. I mean, it's readable this time again. And then we have this nice little see more at the bottom, add to cart.
04:29Alright. Oh, nice. Nice little nice little drawer open there.
04:32And this cart here kinda looks like a trash can though. It's a little confusing, but everything is readable. This is a surprise.
04:38Okay. Navbar. Alright.
04:40That's just terrible UX. I I don't know why it thinks that selecting these. The navbar like that, whatever.
04:45Okay. Our story. Alright.
04:46Not bad. It was better than the last dynamic workflows demo. However, let's see how Fable five matches up.
04:53So I got that link from cursor here. Let me just open this up. Still that funky font, but way cleaner.
04:59We've got this nice little I don't know what you call this little notification thing at the top. It's a classic ecom thing. I've got these nice flavors, guess, kind of rotating, much better buttons.
05:09We have this nice gradient here in the background. The header image calls out the candle like the CTAs as a whole. Yeah.
05:16This is just much cleaner of a header at least. Interesting. Yeah.
05:21Even these images, it's it's it's funny how the, like, model is smart enough to know how to prompt the image generation model, in this case, GBT image two. Like, I didn't give any prompting on how to prompt the model, but it did a way better job of capturing this sort of candle aesthetic type shoot with these little, you know, these little branches and shadows and sun and textured backgrounds, all that kind of stuff.
05:44Even the colors of the candles looks better, and then it just nice subtle little zoom in here on these. Then we have a nice view all candles. Okay.
05:52This is nice too. Shop by ritual, morning rituals, deep work, unwind, what happens? Okay.
05:57Cool. So that's way better filters here as a whole. Nice little product call out.
06:03Go here. Nice. Looks good.
06:05Very similar drawer. Nice little step one, two, three. Here's how to use a candle case you didn't realize.
06:11A nice little testimonial here, email. Oh, I like that. I like that footer.
06:15That's a nice touch. Okay. Yeah.
06:17I mean, I think you see the difference here. Our story. Okay.
06:20Let's go to view all candles. Oh, and then a nice beautiful little three column grid here with all of the different candles. I mean, just nice touch.
06:27So it like took this color from the candle and it knew to, you know, have something like with a similar color, just image prompting, really well done. Looking good. Look at all these.
06:38I mean, again, g p t image two just never seems to amaze me. That's a completely separate aside, but yeah, this is a nice touch. Even like the the little, like, rainy day in the background, like, was it image prompting all of these as it went through one by one?
06:49Like, that is just freaking impressive. Nice filters here. That's not filtered by the flavors, those ugly looking flavors that we had here.
06:58This come on. Come on, Opus. What are you doing?
07:00So I think we have a clear winner. Yeah. You can just see how Fable five is going extra mile and even the little stuff, the tiny details like image generation, like filtering the categories, the nav bar at the top here, even this little notification here, it just kind of matched like a ecommerce type feeling site.
07:17So I think Fable five is the clear winner. Now how much would this have cost me in usage based pricing? So if we go to our tracker here, we can see that Opus $4.08, 198,000 output tokens estimated $21.41.
07:32And then Fable five, a 188 k. Output tokens, so a little bit less, more token efficient, $36.84 equivalent.
07:40Not bad for a entire ecommerce site. I mean, that's like what? One month in a Shopify bill?
07:46And then speed wise, Fable five was probably fifteen minutes faster than Opus four eight, so thirty five ish minutes. Whereas Opus four eight was like closer to 50. So Fable five, at least for this build, despite what people are saying, seems to be faster than four eight.
08:01So that's one point of the board, but one build, of course, doesn't settle everything just yet. Not to mention I'm having way too much fun with these builds. Anyway, so this next one I've been trying to build for literally years and no model has ever come close to pulling it off.
08:13So let's go back into Cursor. I'm just like in the ecom build. We have Fable on the left and Opus on the right.
08:18And what we're building is an interactive three d museum of art history. I am a total art history nerd. And one thing that's always bothered me is if you wanna actually understand the time period, the art pieces that are part of that, your options are basically a textbook or some kind of Wikipedia rabbit hole.
08:33There's no real clear visualization for this. So what I wanna build is a zoomable timeline of every major art period. So you can zoom in to say Baroque and then the famous artists of that era show up.
08:44You can click on Caravaggio for example, and you get this museum placard about him, and then you can actually look at his paintings in a full kind of three d Caravaggio gallery. We're gonna use the Wikimedia comments API to actually pull all of this from Wikipedia, and I'm gonna be extracting it into a database, in this case, Neon.
08:59So this tests a bunch of things at once. First, architecting something this complex from scratch. Second, the actual data side, so extracting hundreds of paintings from Wikipedia, storing them, organizing them.
09:09And then the third, this is the big one, the actual three d world itself and how realistic these models can make it. Because typically, you'll have an output that is just blobby, kind of ugly three JS, but I really want it to look realistic and have shading and lighting and have it be beautiful, like just an immersive experience across the board.
09:25So here is the prompt. Build me an interactive three d art museum of art history that runs in the browser, the timeline, basically everything that I explained, infinite canvas timeline, you go into different artists, has all the different periods, museum view, data all from Wikipedia API, visual quality, and I just emphasize this cannot be a lazy three JS demo.
09:43Make the galleries as realistic as possible, lighting, tone mapping, all of that. Fable five is already asking me about the design direction, but I told it to just come up with three design directions. So I'm gonna say, like I said, just do all three.
09:55Opus is asking me the same thing like I told Fable five. I'm just gonna copy exactly what I told Fable five. Already ran into an issue that Opus four eight called out, public domain constraint versus modern art.
10:05So things like Picasso, Daldi, etcetera, they're not in public domain, so we can't actually get their images. Surprised we can't just take these from Wikipedia. I'm actually gonna push back on that because I think this is just a free app.
10:16I don't see why it can't just include those as images. So apparently, is a copyright thing. Claude's refusing to extract these images, which is fine.
10:24We'll just use the images that are in the public domain for now. So interestingly, Fable five is looks to be done with its output faster than OPUS four eight.
10:32Fable five is testing in the Claw Chrome extension right now, which I'm not looking at, I promise, and it's going through, like, zoom scroll wheels. This is actually something I've ran into quite a bit doing this exact build in the past where the actual zoom, the pan, the movement is is terrible. So it's good that it's actually finding those issues already.
10:49And then Opus four eight, still just building this out. I can just see a lot of things that Fable five is picking up on that is just really complex. This is like needle in the haystack type stuff.
10:58Francis Bacon failed because his article title is Francis Bacon, parentheses artist. So the derived category name is wrong. I don't know even know what that means, but the fact that it's going through all of these different artists, there's probably hundreds of artists here.
11:08It's finding these errors. Oh, I see what happened because in Wikipedia, there's also probably another Francis Bacon that's not an artist. So the parenthesis artist is throwing off the entire category as a whole.
11:17Pretty darn impressive. Another thing I noticed too with Fable five is it's saying, still editing. Meanwhile, the pitch black floor in the museum screenshot bothers me.
11:25Let me look at the floor and lighting settings. Like, it is really quite opinionated even with design, which that is the first time I've ever seen something like that. Fable five two finished probably fifteen minutes before Claude, and it's just been queuing its work the entire time.
11:38Status check so far on Fable five, sixteen periods, sixteen hours, 767 paintings in our NEO database. We have a few timeline views for me to react to.
11:47I'm not gonna look at those yet, and then it's also working on the museum view as a whole. It's also doing a good job of just, like, coordinating agents. So it looks like it has a bunch of different agents here.
11:55It's saying that agent will reinvoke me when it finishes. I'll continue then. So it's just working on stuff in the meantime on its own and then spawning parallel agents.
12:03So it seems to be a pretty good agent orchestrator as a whole. Meanwhile, we've got opus four eight still humming through this build. Okay.
12:10Infable five is officially finished. Let's take a look at it in a second. I'm gonna give Opus four eight a second to catch up.
12:16It looks like it's QA ing its work right now, but that took probably, I'd say, thirty minutes. I'm also almost out of my usage already. I started at, like, 65% usage on the max 20 x subscription plan.
12:28So between the two of these, it burned through almost 40% of my usage in, like, thirty minutes. That's pretty wild. So literally about twenty minutes later, Claude or Opus 40 is running into the same issue that Fable five had twenty minutes ago, where it's this pointer moving around the actual space.
12:44So we'll see how it navigates this. It is actually quite interesting that Fable five is that much faster than 4.8. That definitely did not seem to be the case in all of the demos I've seen.
12:54You can see it's been coding for forty three minutes now. Alright. Let's check out Opus four eight's three d museum first.
13:01Art history, zoom through nine centuries. Enter the Atlas. Okay.
13:06Okay. We've got a we've got a timeline view. I actually like how it color coded this.
13:11Got some nice overlapping with early Renaissance, Northern Renaissance. Is this a proper canvas?
13:17Can I zoom in? Oh, I can. Look at that.
13:18Oh, nice. So they, like, kinda pop up. Ba boom.
13:21Ba boom. It's a little scattered. They're kinda bleeding into each other.
13:24But this is not bad. Not bad at all. Oopus.
13:27Okay. That's where it gets messy. I click in.
13:30No. No. It just isn't working.
13:32I can't click in anymore. Okay. For some okay.
13:34I think it's getting the canvases like canvas movement is overlapping the actual click.
13:40So I can't actually click into these people. View is not bad. I just can't get into the galleries which is kind of annoying that it's broken.
13:48Agent should have QA'd this. These are kinda messy too. Would have been nice to get into the galleries.
13:52We have some other views here. These are just like different types of categories and art movements. This cosmos, I see where it's going with this.
14:00It's a little messy. So I'm in Opus for it. This is probably the best output I've seen.
14:03I've tried this on other models. This is probably when it comes to actual organization and a visual hierarchy.
14:09It did a good job of just me telling it, you know, create a timeline and add artists that way. So that is Opus four eight. Kinda failed a little bit because we didn't actually see the museum gallery, but let's see if Fable five did any better.
14:22So we'll go to five here. View. And k.
14:25I'm liking where this is going. So it did kind of that, like, astrological motif, but a much better job than Opus.
14:33So it's just like it has a nice little timeline. Let me zoom in and like the stars are, you know, people and they kind of like overlap depending on what their when they lived. Pretty sweet.
14:42So you can like zoom in. You could see all the different artists that are featured, you know, in the Northern Renaissance. If you zoom out enough, they like change to little dots.
14:49It's a nice little touch. And then as you zoom in, there's no like overlap between all these people. And it's doing some sort of relation.
14:55I don't know what it is, but these are like connecting to themselves. I'm not sure if that was intentional, if that's just an accident. So really clean.
15:02This is actually exactly what I pictured when I pictured like this moving kind of zoom in infinite canvas timeline. So let's actually go to an artist now. Let's go, uh, let's go dig a nice little click animation there.
15:14We got a little GSAP action. I did request some GSAP. Let's click it one more time.
15:19It's cool. You can like, you know, you can like learn about these. This is who Renoir was.
15:23This is Manet. Let's go into Degas. So Degas and Manet technically should be connected if we're we're being picky because they were friends in real life.
15:33Little fun fact. For that aside, let's go into Degas. Okay.
15:36Here's the gallery. I'm I'm liking what I'm seeing already. Click to enter.
15:40Very cool. Very freaking cool. Wow.
15:43This is awesome. Oh my goodness. So you can move around.
15:46It just just the left and right arrow here and I can move around with my mouse here to actually look at these paintings. I mean, look how cool this is. And then I can walk and look.
15:55This is incredible. They gotta love painting ballerinas by the way and horses. Big ballerina horses guy.
16:01Uh, this is incredible. I mean, imagine having this as a kid. How much more fun would you have learning if you're able to do something like this?
16:08You can go in, you can look at this, you can click into it, we can learn all about this. I mean, this is incredible. This is so freaking cool.
16:15And this is all just from Wikipedia. And it did this across hundreds of artists. Like what?
16:20A thousand painting? Almost a thousand paintings? Fable five, no surprise.
16:24Actually, I am pretty surprised because it's the first time I've seen a model that actually is able to achieve this level of detail and interactivity. The motions are really fluid. We have all the paintings just extracted everything so easily in, like, literally, like, what, thirty minutes.
16:38So definitely have to give this to Fable five considering Opus four eight. I couldn't actually click into the gallery to actually see the gallery. I am curious how Opus four eight designed the gallery, but those are rules.
16:47We only gave Opus four eight one shot. So let's now just look over at costs. Refresh this.
16:53It's four eight, 51,000 input tokens, 437,000 output tokens, estimated cost about $46. Then Fable five, fifty four thousand input, 280,000 output.
17:03Interesting. So it's less output tokens. I wonder if that's right.
17:06Yeah. So it's much more token efficient. I mean, that's what Anthropic said in their announcement, but about 37% more expensive.
17:11So if you think about the actual usage based pricing here, comes into effect in a couple weeks, you're gonna be paying like $64 for this kind of build. Is it worth it?
17:19Probably not. Anyway, moving on. So in my Fable five announcement video, I had it one shot a roller coaster tycoon style game, and it actually worked, which was really impressive.
17:31But the game, it was it was two d if I'm being picky. It was a little too pixelated for my liking. So instead, we're gonna put it to the test with a favorite of mine from my childhood, Age of Empires.
17:41So this thing's gonna be a whole new beast. We're gonna have a full three d world. We're gonna have dozens of units.
17:45We're gonna have enemies that attack you, a whole kind of, like, building the civilization component. I highly doubt it will get even close to this, but we're gonna put it to the test and see both with Fable five and Opus four eight. So we'll open cursor, Fable five on the left, Opus four eight on the right, and paste in this prompt.
18:01Build me a fully playable Age of Empires style real time strategy game in the browser. Core loop all in a full three d world, three JS or WebGL equivalent. I start with the town center, make it look like a real game, playable lighting, and then it's already asking which visual style?
18:15Stylize a low poly recommended. No. I don't want that.
18:17I want it real. Painterly realistic. Multi texture.
18:21I honestly have no idea. It's funny. It's the exact same questions.
18:24Okay. So that only took about thirty minutes. Thirty three minutes for Opus and thirty minutes for Fable five.
18:30Now let's see which one is better. Opus first, Bronzedon, Forging the World.
18:36Okay. Already not even close to Age of Empire graphics. I mean, are just blobs.
18:42Can I move around? No. Left click select.
18:45Every time I do these game demos, don't know how to play these things. It's just so confusing. Not intuitive.
18:50I almost need Opus to just show me how to do it. Left click select. Okay.
18:54I can select town center, bottom left here, right click move, gather attack, rally point set. Okay. Not sure what this that means.
19:02House. Can't build there. There's not like a town center.
19:05Can't zoom in. There's no movement. I think this app is not working.
19:10Yeah. Okay. I'm not gonna waste any more time on this.
19:11It looks like the completely broken. I can't even move like around the world. So Opus did not complete the task.
19:19So let's see if Fable five is any better. Again, my expectations are relatively low. I know this is hard.
19:25We'll go to Fable five's demo here. Empires of Dawn, begin your reign. Select with left click.
19:31Drag command with right click. Okay. I'll figure it out.
19:33Okay. Now we're talking. Woah.
19:36Alright. Map moves. There's our civilization.
19:38This is insane. This is I mean, this is exactly what the, like, graphics were. Like, this is just as good as Age of Empires.
19:44This is wild. I mean, look at the freaking difference. Like, is like, this is not a single model upgrade just in terms of graphics.
19:51Nothing else. I don't know what the actual gameplay is and I'm probably not gonna get too much of the gameplay because I never know how to play these actual games. But like, look at this freaking thing.
19:58This is wild. So okay. Alright.
20:00Let's just try to click some buttons here for idle. Okay. House.
20:04Oh, nice. You can oh, wow. That's cool.
20:06Look at this. Oh, my god. Alright.
20:09Farm. Left click to place. Right.
20:11That's and then it looks like we have, I don't know, some things over here, and then we have some these are the enemies. Look at this. They already have they already have armor.
20:18I mean, I just cannot get over these graphics. This is insane. Enemy town center.
20:22Do we attack them? This, by the way, is gonna be available on Steam, February. Okay.
20:26What do we do here? I don't know. Less five population.
20:28I don't know how to actually play these games, but it looks like it's fully playable. And, like, the map is navigable. Okay.
20:33Not enough resources, I need it. I need resources? This.
20:37Wow. Uh, Yeah. It just absolutely crushed it.
20:40Like, I knew it was good at three d. I've seen some game demos of Fable five. They're really good, but, like, this is this is just on another level, especially compared to opus four eight.
20:50I will let you be the judge, but I think we've got a clear winner here. I, of course, don't love this though because we will now be getting used to a certain level of quality that just cannot be matched by opus four eight, which is surely Anthropics plan all along. Anyway, let me know what I should be experimenting with next.
21:04If you made it this far, I hope you had fun watching along. Be sure to leave a like, subscribe, helps out a ton. Now get back to building.
21:09We've got less than two weeks of
The Hook

The bait, then the rug-pull.

Three builds. One prompt each. No revisions. Pat Simmons gave identical instructions to Opus 4.8 and Fable 5 simultaneously inside Cursor, deployed every output live, and filmed the results -- and the gap was, in his words, not even close.

Frameworks

Named ideas worth stealing.

00:00model

One-shot build comparison methodology

  1. Same prompt to both models
  2. No revisions after initial output
  3. Deploy live
  4. Measure: quality, speed, cost, token efficiency

A reproducible framework for comparing AI coding models on real-world agentic tasks rather than benchmarks.

Steal forEvaluating LLM upgrades for production coding workflows
CTA Breakdown

How they asked for the click.

VERBAL ASK
20:55subscribe
Let me know what I should be experimenting with next. Be sure to leave a like, subscribe, helps out a ton.

Soft and brief -- comes after strong genuine enthusiasm so it lands naturally.

MENTIONED ON CAMERA
FROM THE DESCRIPTION
Storyboard

Visual structure at a glance.

open
hookopen00:00
Build 01 title card
promiseBuild 01 title card00:23
Opus ecom live
valueOpus ecom live02:47
Fable 5 ecom live
valueFable 5 ecom live04:39
Build 02 title card
promiseBuild 02 title card08:01
Opus museum broken
valueOpus museum broken13:05
Fable 5 museum gallery
valueFable 5 museum gallery15:00
Build 03 title card
promiseBuild 03 title card17:23
Fable 5 game playable
valueFable 5 game playable19:15
Closing
ctaClosing20:49
Frame Gallery

Visual moments.

Watch next

More from this channel + related breakdowns.

Chat about this