The argument in one line.
When the same marketing prompts run on two AI models, the one that reliably finishes a task end-to-end beats the one that occasionally looks nicer but stalls unfinished.
Read if. Skip if.
READ IF YOU ARE…
- A creator or marketer deciding which AI model to lean on for day-to-day production work like websites, thumbnails, and clip-mining.
- Someone running the same prompt across multiple models and wants a real-world example of how to judge the results beyond 'which one looks nicer.'
- A solo operator trying to figure out how much marketing production work can now be handed to an AI end-to-end.
SKIP IF…
- You're looking for a technical benchmark or reproducible eval — this is one person's informal, single-run comparison, not a controlled test.
- You want a step-by-step tutorial on any one tool (Higgsfield, Riverside, Single Brain) — none are explained in depth here.
The full version, fast.
Eric Siu runs the same four marketing tasks — building an interactive website world, redesigning YouTube thumbnails, writing an Amazon growth-strategy memo, and auto-clipping a webinar for social — on both GPT Sol 5.6 and Claude Fable 5, using identical prompts each time. Sol 5.6 wins consistently, not because its raw output always looks better, but because it finishes each task end-to-end with far less back-and-forth: one prompt yields a working website world in about 35 minutes, thumbnail concepts land close to what he actually ships, and webinar clips render without hand-holding. Fable 5 sometimes produces a prettier intermediate result but stalls, fails to fetch a source page, or needs repeated iteration to finish. His conclusion: reliability to completion currently beats aesthetic polish for practical marketing production.
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 →Where the time goes.
00:00 – 00:25
01 · Cold open: the tease
Eric opens by declaring Sol 5.6 ‘crazy’ for marketing and better than Fable 5 in many cases, teasing the head-to-head to come.
00:25 – 02:33
02 · Test 1: SingleBrain.com built into an interactive world
One simple prompt on Sol 5.6 turns the SingleBrain.com homepage into a full explorable ‘world,’ running unattended for roughly 35 minutes; Eric shows the exact one-line prompt and the token spend.
02:33 – 06:16
03 · Test 2 & 3: YouTube thumbnail identity and the Amazon growth memo
Sol 5.6 is fed the transcript of Eric's prior video and iterates on new YouTube thumbnail concepts for the Leveling Up channel; separately it produces an Amazon growth-strategy memo allocating investment across AWS AI, Ads, and supply chain.
06:16 – 06:48
04 · Sponsor break: Single Brain
Eric plugs Single Brain, the managed-revenue-agent product behind the SingleBrain.com site being world-built in test 1.
06:48 – 09:16
05 · The rematch: Fable 5 attempts the same tasks
Running the identical thumbnail prompt on Fable 5 fails outright when it can't download the source channel page and burns Higgsfield credits; the SingleBrain world-build on Fable 5 looks visually stronger but is still unfinished three hours in, while Sol 5.6 shipped in well under an hour; Eric also revisits a growth audit that leaned on Amazon's grocery network, calling it a miss.
09:16 – 11:16
06 · Test 4: auto-clipping a webinar for social
Eric asks each model to mine a recorded webinar for short, mid, and long-form social clips with strong hooks; one model stalls through repeated retries while the other lays out a full deliverables plan before being told to actually render.
11:16 – 13:20
07 · Reviewing the finished clips and the final verdict
Eric plays back the AI-selected clips, flags what still needs manual audio and pacing work, and delivers his overall verdict: Sol 5.6 wins on early-stage marketing work because it finishes reliably.
Lines worth screenshotting.
- Sol 5.6 turned a flat marketing homepage into a full interactive 'world' from a single prompt, running unattended for roughly 35 minutes.
- Running the identical thumbnail-design prompt on Fable 5 failed outright because it couldn't download the source channel page — wasting Higgsfield credits before doing any real work.
- Sol 5.6 finished the YouTube thumbnail task in one shot with minimal back-and-forth; Fable 5 needed several rounds of iteration to reach a usable design.
- On the website world-build task, Fable 5's visual concept looked better, but it still hadn't finished after three hours, while Sol 5.6 shipped its version in well under an hour.
- Sol 5.6's Amazon growth-strategy memo allocated roughly 80% of incremental investment to AWS AI infrastructure, splitting the rest between Amazon Ads and Amazon Supply Chain Services.
- The two models' Amazon growth audits disagreed: one leaned on Amazon's grocery network as the growth lever, which Eric calls a miss — he'd rather see the investment go toward supply-chain services.
- Asked to auto-clip a webinar into short, mid, and long-form social cuts, one model stalled through repeated retries and only finished 10 of 21 requested clips.
- The other model produced a full deliverables plan — formats, hooks, camera treatment — before actually rendering anything, and had to be explicitly told to finish.
- Riverside's own network access policy blocked one model from pulling the source recording directly, an infrastructure snag that had nothing to do with the model's actual capability.
- Even the AI-selected clips that nailed the hook and topic still needed manual audio cleanup (room echo) and stronger mid-clip payoff before they'd be publish-ready.
- Eric's standing verdict, repeated across all four tasks: the model that finishes reliably with the least hand-holding wins, even when a competitor's raw output occasionally looks nicer.
Finish-power beats pretty in every single test.
MODEL COMPARISON
Across four real marketing tasks, the model that completed the job end-to-end with the least hand-holding won every time, even when its output wasn't the best-looking one on screen.
02Test 1: SingleBrain.com built into an interactive world
- Turning a flat marketing page into an interactive, explorable 'world' is now a single-prompt task, not a design sprint — the AI ran unattended for roughly 35 minutes and returned a full explorable build.
- A model that completes a build end-to-end without needing constant re-prompting is worth more in practice than one that drafts a slightly nicer first pass and then stalls.
- Even a 'successful' one-prompt build should be treated as a rough draft — expect to go back and refine the result before it's actually usable.
03Test 2 & 3: YouTube thumbnail identity and the Amazon growth memo
- Feeding a video transcript into an AI and asking for thumbnail concepts can turn a design task into a few minutes of back-and-forth that lands close to what you'd actually ship.
- A strategy memo is a good stress test for a model's business judgment, not just its writing — two models can disagree on where to place a company's investment, and only one may be right by your own domain knowledge.
- Asking for a percentage-allocation breakdown forces a model to commit to a specific, checkable answer instead of vague advice.
05The rematch: Fable 5 attempts the same tasks
- Re-running the exact same prompt on a second model is the cleanest way to compare them — it isolates model behavior instead of prompt quality.
- A model that fails to pull a required input can burn through third-party tool credits before doing any real work — a hidden cost that doesn't show up until you check the credit meter.
- Prettier intermediate output isn't the same as a finished deliverable — a visually stronger concept that's still unfinished after hours loses to a plainer one that shipped in under an hour.
06Test 4: auto-clipping a webinar for social
- Clip-mining a long recording for short/mid/long-form cuts is now something you can hand to an AI wholesale, with instructions as specific as clip length, hook strength, and platform aspect ratio.
- A tool that requires the AI to re-fetch source video from a third-party platform introduces a failure point outside the model's control — check where your source lives before trusting an end-to-end clip pipeline.
- Plan quality and finish reliability are two separate axes to judge — one model can produce a great plan and never render it, while another renders reliably without a fancy plan.
07Reviewing the finished clips and the final verdict
- AI-selected clips can nail the hook and topic selection while still needing manual audio cleanup and pacing fixes — don't expect a publish-ready cut on the first pass.
- A broader, less conversion-focused clip can still be worth publishing for reach and audience-building, even if it won't drive direct conversions.
- The reviewer's standing rule — pick the model that finishes reliably over the one that occasionally looks nicer — held across all four tasks tested, not just one.
Terms worth knowing.
- World-building (AI context)
- Turning a flat webpage or brand into an interactive, explorable visual environment — more like a 3D scene you can move through than a static page — generated by an AI model from a single prompt.
- Higgsfield
- A CLI and platform used to generate and preview interactive 'world' builds from a website, billed on its own separate credit system.
- Managed revenue agents
- Single Brain's product category — persistent AI agents assigned to a client account that carry context across outbound, follow-up, and reporting instead of starting fresh each time.
- ASCS (Amazon Supply Chain Services)
- A newer Amazon business line focused on supply-chain logistics services, referenced as a growth-investment target in the video's AI-generated Amazon strategy memo.
- Riverside
- A remote recording platform used to capture the webinar footage that one AI model was asked to pull and clip for social media.
Things they pointed at.
Lines you could clip.
07:28
“So I would say by a mile, 5.6 wins when it comes to marketing on the YouTube thumbnails piece.”
07:43
“It failed to download the leveling up channel's page. That's gonna affect things a lot.”
12:04
“The three ways that men go broke: lays, liquor, and leverage.”
13:02
“Your mileage may vary, but as of this video, when it comes to early marketing, 5.6 is the winner.”
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.
metaphoranalogy
00:00GPT solve 5.6 is crazy when it comes to marketing. Now, I've been comparing it versus Fable five and have another video for that that you can check out.
00:08But I wanted to show you this video as it pertains to marketing, and I'm even gonna do a comparison of how solve 5.6 stacks up. I'm telling you right now based on what I'm seeing, it is crazy. It is insane, and I would say it's even better than Fable five in many cases.
00:22So let's check it out. So check this out. Sol 5.6 made me this new website.
00:27So this is a single brain new website. I'm scrolling, and it's it's created as new world over here.
00:32Okay? So this new world over here By the way, if you wanna see the old website here, I'll I'll just show the old website real quick. So if you look at the old website, this is what the old website is.
00:40So this is managed revenue agents that know the client. Right? And this is pretty good.
00:44It it gets to the point, and you can see how it works over here. This is what we sell for for clients, these managed revenue agents. And so this is not bad, and you know, obviously this is generated via AI, but this version, now we're starting to get somewhere.
00:56And it it probably took me like the the prompt was very basic, and I'm gonna show you the prompt in a moment. But it took me, what, five minutes to just get this going, and then maybe I don't know. It it looks like this this worked for about thirty five minutes or so, and I just the prompt was actually very simple.
01:09So again, every signal is still disconnected. You can see it. It basically made this world, and you can scroll through it.
01:15So I have a lot that I can play with now. Obviously, there's a lot more that I need to massage here, because again, only put one prompt here. And let me show you what that one prompt looks like.
01:23So let's go to g p d solve 5.6, and let's take a look at this real quick. So this is my original prompt. Install the skill and build singlebrain.com into a world.
01:32All I did was I took a tweet. Okay? And then it it did some work over here.
01:36It worked for about fifteen minutes, and I'm using 5.6 sol high, and you can see okay. So I installed so I need to install the the Higgs field CLI. So I I handled all that, and I said, okay, I'm in.
01:45So after I handled that piece, it asked me for approval, I did work for about sixty minutes or so, and then it showed it showed me kind of the concept with this world. So here's the world concept, and there's probably a lot more that I can do with this, but this is pretty cool. Because the whole idea of having a single brain is that if you have a single brain as as your main brain, but then your teams have brains as well, so you have a team brain, you could have an individual brain as well, and then it just just goes out levels and levels.
02:09Right? So this is pretty cool what it drew out here, these different world concepts. I didn't tell it to do this.
02:14It kinda just came up with this. It it just assumed that with little direction it came up with this. Right?
02:19So then it worked for another thirty five minutes or so, and then what I just showed you was the interactive world, site that I showed you where it has that scrolling the scrolling ability. And so that gives me something to play with, and I'm just gonna continue to work with this, and I spent about $504,167 credits, about 766,000 tokens over thirty four minutes or so.
02:35And you know, honestly, I'm I'm curious to see what it'll look like if I I move this over to ultra. So so that's that. And that's just one piece of now, I'm gonna give you one more piece.
02:43So the next thing I did, I just made a video right before this one, and I took the transcript from it, and I said, hey, want you to look at the thumbnails for my leveling up YouTube channel to determine our thumbnail identity. Give me three thumbnail and headline options based on this video transcript. So it works for six minutes, and then it comes up with the ideas over here.
02:59So here are the ideas, and I said, okay. We'll design out the actual YouTube thumbnails. And so this is not bad.
03:05This is actually what my thumbnails look like on my YouTube channel. This is not bad. It's it's not a bad start for what this prompt is.
03:10I am using 5.6 ultra here, which is the the highest version that you can see. And then I I then I start giving a feedback a little back and forth here, and then I say reimagine number two. Okay.
03:18Great. So I I keep working with it, and then we end up getting some new versions over here. So I I refine it a little more.
03:23I'm like, hey. This is one that I actually wanna use over here. Sol 5.6 versus Fable five.
03:27And then I have this other one over here on, hey, 5.6 might be cheaper, and then you have this one over here where it seems like it's it's it's locked down a little bit. Right? So it's you kinda decide on the trade offs here.
03:36But I actually think that this this thumbnail is actually not that bad. And so then I realized something. Was like, hey, you know what?
03:40I don't think we're actually using the Fable five butterfly mark. Okay? So let's take a look at this.
03:44So Fable five okay. Now we have the butterflies over here, so that that actually makes sense now. Fable five design.
03:50Okay. You know, maybe we need to have the Claude logo right behind it. Maybe we should do that.
03:55Right? So I don't know. Like, do I wanna use that or do I wanna use this?
03:59Because that that one's more that one's more recognizable. Right? So I don't really know.
04:03Like, you know, I I think this one actually stands out versus this one. I don't know.
04:09But now I have different options, and so that's really good. Now the other thing is I had all these other loops running where I'm I'm making this one for recruiting right now. So we have a recruiting operating system.
04:17This one's for billing for about two days or so, and you know, looks like it's it's kinda held back right now, but I'm running it on 5.6 now, and it's doing a better job of moving it along in terms of specking out the plan more. So when you wanna plan better, you wanna plan deeper, you can you can use this. Right?
04:31So it's it's pretty good for for building a lot of these marketing loops, lot of these like SEO, AEO loops that you might have. And what I would say is, as I continue on here, I wanna show you that this this Amazon growth proposal or these growth bets over here. So this is really good because the prompt originally was, hey, I want you to advise Amazon senior leadership on the three incremental growth initiatives that would create the most enterprise value over the next three years.
04:55Okay. So basically, it does all this this this analysis, and it looks like it worked for about nine minutes here, which is not that bad.
05:02And then versus, I mean, Fable five actually took a little longer, which is okay. But the analysis was pretty good. It says, hey.
05:08You know, let's I would invest in AWS AI utility, Amazon ads, and then ASCS, which is Amazon's new supply chain services. And you can see how much would it allocate to 80% AWS AI, 10 to 15% ads, and then 10 to 15% ASCS.
05:23Okay? Which is pretty good. And then it shows you the revenue for for the different segments that they have, the different analysis, different opportunity ranking.
05:29So I'm showing you all this right now because I'm showing you that this does pretty good financial and business analysis. It did a pretty good job with the website on the on on its first shot, basically.
05:39It did a really good job with the thumbnails ultimately, so that I that tells me I can move a lot faster when it comes to my thumbnails. And thumbnail packaging matters a lot, like a lot when it comes to YouTube, and you really gotta understand your stuff. Now what I can do, I I could just make that into a skill.
05:52In fact, I'm I'm gonna have you watch me make that into a skill right now. This looks like something that'd be amazing for a skill. Let's make this into a skill and then tell me where you would put this so we can store it somewhere.
06:02Right? So let's make it into a skill and then we'll and then we'll go from there. But it did all the analysis kind of, you know, basically mostly end to end here.
06:09Okay? And the reason I say end to end right now is because I'm gonna show you the difference when you take a look at Fable five. By the way, if you're enjoying this video on 5.6 for marketing, then you gotta check out Siegel Brain.
06:19So Siegel Brain is where we build these managed revenue agents for companies where we help them scale up their creatives a lot faster, scale up their outbound a lot faster, scale up their AEO a lot faster. Whatever it is exactly, whatever business problems that you have, go check out singlebrain.com with ab, and we'll see you on the other side.
06:34Okay. So here's the difference now. I ran the same exact prompt on my collage.
06:38I want you to look at the thumbnails, and I I gave the transcript over Literally the same exact prompt. Okay? And it decided to use Higgs field here, which is fine, and then it it used So here's the thing.
06:49It failed to download the leveling up channels page. That that that's gonna affect things a lot. If you can't pull the initial base, you're not gonna have a lot to work off of here.
06:56Right? And so very well, it could could be better, but it I don't wanna spend the time having to worry about the technical issues here. I don't wanna spend time about having to think about how to go about connecting this and and things like that.
07:07Right? So there's some cool concepts here, but the other thing is it's it's eating up my Higgs field credits. I don't want it to eat up my Higgs field credits.
07:13The other one, it did it in one shot. It basically did it in one shot across the board. It didn't ask me for all these other things.
07:17It's not taking my Higgs field credits either. So I would say by a mile, 5.6 wins when it comes to marketing on the YouTube thumbnails piece. Okay?
07:26Let's say I move over to the singlebrain.com world building. Okay.
07:29Here's the thing. It's still building right now. Okay?
07:31Like, I I put the same prompt in here three hours ago. It's still building. Okay.
07:34It came up with these this over here, but it it's it's struggling over and over and over again. It's struggling. It's struggling.
07:39It's struggling. Okay. Here's a world.
07:41Okay. Let's take a look at this. What what does this look like?
07:45Okay. This is kinda cool. Okay.
07:47Here's a brain. Okay. That that's actually really cool, the single brain piece over here.
07:51Right? Because we're talking about a single brain. That's kinda cool.
07:56Alright.
07:59This is my first time seeing this, by the way. So here's another one, and it goes a little deeper. Okay.
08:04Cool. So let's just make this like, I would say the designs here for the the the brain piece of it is is objectively better, but in terms of finishing it off in one shot, solve 5.6 did a better job. And so I I like a lot of these.
08:17I'm just like, okay. Well, damn. Okay.
08:19Well, great. Great. This is great.
08:22So but now show me the preview, the interactive preview. But fact of the matter is I would still give Sol 5.6 more credit here because it I didn't have to do a lot of back and forth.
08:33I just wanted to figure it out and go from there. Okay? Now the other piece is this.
08:37So I did show you the Amazon growth initiatives audit over over here. Okay? The audit that it did over here was it said, hey, it would actually focus on the grocery network that it has, which I think is a miss.
08:49That to me is it's That's that's not a very good audit because I actually would focus on AC, Amazon supply chain services. That that's what it really is at the end of the day.
08:56Now I'm gonna show you one more thing. K? So I wanted it to find me social media clips from this webinar that I did yesterday.
09:03Okay. So I said, hey. I want you to clip this up for social media mid form and short form clips with strong hooks to start.
09:07Mid forms being two to five minutes and short forms being sixty to ninety seconds, ideally added to proper dimensions. Also clip up for YouTube long forms that are 10 each with a montage of highlights from the episode and even some hangers to to that that make people wanna watch.
09:19Okay? So it's going back and forth, back and forth, done, trying again, done, trying again. Okay.
09:23Like, where where are we at with this? Like, what's going on? Where are we at?
09:27Why do we keep failing? Goal is to finish. Oh, no.
09:31No. It's still running right now. 10 of 21 clips.
09:32Okay. Great. So it's rendering right now, but it it yeah.
09:35We had to keep doing this this back and forth thing, which is really annoying to me. Okay? Conversely, if I show you what is happening over here on the GPD side of things, okay, it actually gives me the different ideas that it has.
09:50Okay. So let's make sure this is the right prop. Okay.
09:52So you know, the mid form claims blah blah blah. Okay. Great.
09:55It shows me okay. Here's the deliverables. Okay.
09:56So it it found for short form, here's an idea, here's an idea, here's an idea here for short form. Here's a mid form one, here's a second mid form one, here's a longer form one, and that's basically it. And here's the visual system and all that, the camera treatment, all that type of stuff.
10:08Right? So I told it to just finish it off. Okay.
10:10So then it it this one also didn't give me the edited videos as as well. So this one, let's see if it finished it off. So let's see what it says for video folder.
10:19I'm looking at the videos. Okay. It it actually looks like we do have some videos here.
10:24So so Sol 5.6 actually finished this off, and I pulled from my marketing school that I recorded with Neil yesterday. So check this out.
10:33This is amazing. Alright. So check this out.
10:35It it looks like this is the first time watching this with you. Okay? So let's see if this is actually a good short form video.
10:39I predict 50 of companies will need new leadership because the old management style won't work in the era of AI. That's exactly why I'm seeing 95% of so called AI transformation initiatives fail.
10:50Most are bolting on AI pilots for functional tasks, but never touching the business core. And this is something that we're seeing quite a bit as well, so we should react to this. 50% of companies will need new leadership because the old man should sell.
11:02Okay. So what am I gonna say about this? This is it's a good start, but also, like, it's it's a little choppy in the middle because we're not really giving the the meat or the payoff here.
11:11Obviously, there needs to be some audio treatment for this one too because you can kinda hear a little bit of the of the room. I'll try to look at another one here. So alright.
11:17So this is number two. Let's see how good this is. It works so well that it took double the length to do the freaking new lane if we did all manual.
11:25We hired a company that showed me a video of how in a factory, they use technology to build a frame of your house, and then it just plunge from place and they're bringing it over in truckloads so that way Okay. So this is actually not bad. So if we have some good treatment here, there's some good b roll here, this is actually a video that could do well because also this kind of it's a broader topic too.
11:45So it's gonna get more views and not necessarily gonna drive conversions, but it will drive awareness ultimately. Because Neil here is talking about the the the the home technology that he bought that ended up they said it would take shorter time to build itself, it actually ended up taking longer. So it was actually a pain in the butt.
11:58So this is like there's like a good payoff for this one. I'm gonna do one more here, so give me one second. Alright.
12:04Let's see how good this one is over here. This is me again. Now the three ways that men go broke, lays, liquor, and leverage.
12:10My friend Charlie Munger said this. He yes. And it's really the last one that matters.
12:14It's leverage. He he made up the first two. Laking is a liquor fees is just for fun.
12:18It's the leverage. So it's not saying you don't take out leverage, but there's, like, too much leverage, and there's also bad leverage too. What Neil's also talked about too is leverage is an amplifier.
12:26Right? Just like . So if you have a good business model, leverage will amplify a good business model, but also amplify a bad business model too.
12:35Okay. Guys, I think we have something here. So there's already two immediate things here that I think we could scaleify.
12:40There's the thumbnail piece, and then now I'm like, oh, finding moments with short form, mid form, long form. I feel like short form, finding the right moments, we're talking about taste, that problem is solved. Okay?
12:50The thumbnails piece, that is solved as well. The other things I showed you, building a better website, building a more interactive website, solved as well. And so your mileage may vary, but as of this video, when it comes to early marketing, 5.6 is the winner.
13:03And nothing I don't like Fable five. I love Fable five, by way. There's a lot you can do with it.
13:07Uh, there's some cool stuff that you saw with it, but I choose Chat Cheapity 5.6 with some of the marketing stuff that we're talking about early days.
The Hook
The bait, then the rug-pull.
Eric Siu opens with a blunt claim: GPT Sol 5.6 is ‘crazy’ for marketing work, and in many cases beats Claude Fable 5 outright. What follows is a real side-by-side — the same four prompts, run on both models, judged not on which output looks nicer but on which one actually finishes the job.
Named ideas worth stealing.
12:04list
Three ways men go broke (lays, liquor, leverage)
- Lays (relationships)
- Liquor
- Leverage
A Charlie Munger line quoted inside one of the AI-selected webinar clips — of the three, leverage is framed as the one that actually matters, since it can amplify either a good or a bad business model.
Steal forframing content around 'the one lever that actually matters' out of a list of common failure modes
How they asked for the click.
VERBAL ASK
06:16product
“check out Single Brain — managed revenue agents for companies scaling creatives, outbound, and AEO”
Mid-video plug tied directly to the task he'd just demoed (the SingleBrain.com site being world-built), so the ad reads as a natural continuation rather than an interruption.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
OTHER LINKSAlso linked in the description.
Visual structure at a glance.
