One creator ran the exact same week of real work through both frontier models, a website, a game, a script, an inbox, a full business audit, and tracked where they actually diverged.
Across parallel real-work tests, two frontier models produced comparable creative output, but the sharpest difference showed up in judgment on ambiguous real-world data, where only one model verified stale-looking information instead of taking it at face value.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
You're actively deciding between GPT-6 and Fable 5.1 (or a comparable frontier model) for daily work, not just curious about benchmark scores.
You do 'vibe coding' as a non-technical builder and care which tool needs less terminal hand-holding to finish a working app or game.
You use AI to audit your own business or content files and want to know how much you can trust its conclusions.
You're trying to get an AI to write scripts in your own voice and want to know whether feeding it prior transcripts actually helps.
SKIP IF…
You want a controlled, benchmark-score comparison; this is one person's subjective walkthrough of personal tasks, not a lab test.
You're looking for coding-architecture or API-level analysis; every task here is consumer-facing app and chat usage.
TL;DR
The full version, fast.
A creator ran GPT-6 and Fable 5.1 through the same set of real personal-work tasks over 48 hours: a website redesign, an interactive presentation, two vibe-coded projects (a game and a to-do app), a YouTube script, a video edit, email triage, and a full business audit. The two models traded wins on design and writing, and both struggled to sound human in scripts and made identical editing mistakes. The clearest gap opened in the business audit: one model asked for missing data before concluding, caught a stale invoice record by checking Gmail directly, and factored in an unannounced platform change; the other treated the file folder as ground truth. The bottom line: comparable quality, but one model costs less for the same 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.
States the premise: both models tested side by side on the creator's real workflows over 48 hours.
00:26 – 04:21
02 · Website redesign test
Both models redesign the creator's site from reference material only. GPT-6's version works fully but plays it safe; Fable 5.1's version looks better on-brand but visibly borrows structure from the original despite instructions not to, and has some broken UI elements.
04:21 – 06:41
03 · Presentation / design test
Both models rebuild a scrollable, interactive presentation site. The one built inside a dedicated design tool feels smoother and more polished than the chat-app version built with identical instructions.
06:41 – 09:32
04 · Vibe coding: game test
Both models are asked to build an Animal Crossing-style game for fun. One model's game one-shots working mechanics but looks small; the other looks nicer visually but the mechanics needed more digging to unlock (like decorating a home).
09:32 – 13:17
05 · Vibe coding: real app + non-technical usability
Both models build a voice-to-task-list app end to end. Functionality is close, but one model installs its own dependencies and launches the finished result, while the other requires the user to run terminal commands and open other apps manually.
13:33 – 18:21
06 · YouTube script writing test
Both models write a script cold, then again with the creator's prior transcripts as style reference. Both sound noticeably more AI-generated on the first pass; feeding in past transcripts helps but doesn't fully fix stock AI phrasing.
18:21 – 19:58
07 · Video editing test
Both models edit raw footage using an MCP connection plus computer use to inspect the timeline. Both miss the same false starts and overuse zoom effects, but one model also blurs an exposed API key on its own, unprompted.
19:58 – 20:33
08 · Email triage test
Both models triage an inbox and draft replies without sending them. Both perform equally well, undercutting the case for paying for the top-tier model on this specific task.
20:33 – 23:51
09 · Business audit / knowledge work test
Both models are given a folder covering the creator's whole business and asked to audit it and recommend next steps. One model asks for missing analytics data, catches a stale bookkeeping record by checking Gmail directly, and factors in an unannounced platform change; the other treats the folder as ground truth and misses all three.
23:51 – 25:15
10 · Verdict
The creator's bottom line: comparable output quality overall, a meaningfully better audit from one model, and a lower price for the same or better results.
Atomic Insights
Lines worth screenshotting.
A test that handed both models a business folder and asked for a full audit surfaced the sharpest gap between them: only one model asked for missing data before drawing conclusions.
One model caught that a bookkeeping record showing an unpaid sponsor invoice was stale, then verified in Gmail that the payment had actually already landed instead of trusting the ledger.
In a non-technical vibe-coding test, one model installed its own dependencies (a 3D engine and a game engine) and launched the finished app; the other asked the user to run terminal commands and open apps manually.
The model that 'won' the aesthetic comparison on a website redesign had also visibly copied structural elements from the reference site it was explicitly told not to copy.
AI-written YouTube scripts still read as AI-generated even after the model was given a creator's own prior transcripts as style examples, though the extra context measurably reduced generic phrasing.
For simple, well-defined tasks like inbox triage, both frontier models performed equally well, meaning the more expensive model bought no better outcome for that job.
A video-editing test that let a model inspect a timeline via computer use resulted in it automatically blurring an exposed API key on screen, without being asked to.
Both models made identical editing mistakes on the same footage (missed false starts, overused zoom effects), suggesting automated rough-cut editing quality has plateaued across current frontier models.
Functional completeness and visual polish moved independently across every creative task tested: a model could win one axis while losing the other.
The final verdict came down to price more than raw capability: one model delivered comparable or better results at a lower cost per task.
Takeaway
Judgment on ambiguous data is the real gap
WHAT TO LEARN
When two frontier models produce similar-looking creative work, the real difference between them shows up in how each one handles messy, real-world data with no single correct answer.
02Website redesign test
A model can technically follow an instruction like 'don't copy the reference' while still visually resembling it, so verify that kind of constraint by comparing outputs side by side rather than asking the model if it complied.
Functional completeness (every button working) and aesthetic appeal are separate axes of quality, and a model can win one while losing the other on the exact same task.
03Presentation / design test
A design tool built specifically for design work can still outperform a general chat model given identical instructions, because interactivity and polish are easier to produce in a purpose-built environment.
Vague creative instructions produce blander results than detailed, specific ones across every model tested, and that gap matters more than which model is used.
04Vibe coding: game test
A model that one-shots a fully working build with zero follow-up fixes is a meaningfully different outcome from one that is merely 'mechanically fine,' even when both are technically playable or usable.
Visual polish and functional mechanics can diverge sharply between models on the same prompt, so judge the two separately instead of averaging them into one impression.
05Vibe coding: real app + non-technical usability
For a non-technical user, the deciding factor in a coding tool often isn't code quality, it's whether the tool installs its own dependencies and launches the result versus leaving the user to run terminal commands.
A model that can install and configure other applications end to end removes an entire category of setup friction for people who don't want to touch a terminal.
06YouTube script writing test
Feeding a model prior examples of your own writing measurably softens AI-sounding phrasing, but it doesn't eliminate it, so plan to still hand-edit the output.
Certain stock phrases read as AI-generated regardless of which frontier model produces them, making them a reliable pattern to search-and-destroy in any AI draft.
07Video editing test
A model that can inspect a screen directly, rather than working purely from an API, can catch things a code-only workflow misses, like an exposed key that needs blurring.
Two models given an identical editing task made identical mistakes, which suggests a shared limitation across current frontier models rather than a weakness specific to one of them.
08Email triage test
Simple, well-defined tasks like inbox triage are already commoditized: the more expensive model bought no better outcome than a cheaper model likely would have.
09Business audit / knowledge work test
A model that treats a folder of files as ground truth without question will repeat stale or wrong records, while a model that treats the same data skeptically will go verify it against the source.
The single biggest differentiator between two similarly capable models wasn't writing or design, it was whether the model proactively asked for missing information it needed before drawing a conclusion.
Real-world knowledge work rewards a model that notices context nobody explicitly supplied, like a platform quietly changing how it counts something, over a model that only answers the literal question asked.
Glossary
Terms worth knowing.
Vibe coding
Building an app or game by describing what you want in plain language and letting the model write, run, and iterate on the code, without the user reading or editing the code directly.
Computer use
A model capability where the AI operates a computer directly, moving a mouse, clicking, and reading the screen, instead of only calling an API or writing code.
MCP connection
A Model Context Protocol link that lets an AI model control a specific app (like a video editor) through a defined set of actions, rather than through raw screen control.
Resources
Things they pointed at.
00:00productGPT-6 Astra
00:00productFable 5.1
04:21toolClaude Design
06:41toolBlender
06:41toolGodot
09:32toolXcode
Quotables
Lines you could clip.
10:53
“All that said, I think the winner for design work, even if it's just by a little bit, is Fable 5.1.”
clean verdict line on a subjective creative test→ TikTok hook↗ Tweet quote
13:17
“For a non-technical user, if you want to vibe code anything, build some games, some apps, I think GPT-6 is much easier to use.”
“The way that GPT-6 approached this feels to me like the closest thing I have ever seen to AGI.”
hyperbolic, quotable claim tied to a specific concrete test→ TikTok hook↗ Tweet quote
22:49
“Fable 5.1 also told me that one of the sponsors I'd worked with months ago never paid me because my bookkeeping records said they still owed me money. GPT-6 saw that as well, but then on its own, it went out, did some investigation by going into Gmail and realized I was actually paid.”
concrete anecdote that proves the judgment gap rather than asserting it→ newsletter pull-quote↗ Tweet quote
24:31
“You're going to get a lot more for the same price with GPT-6 than you're going to get with Fable 5.1.”
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
metaphoranalogy
Well, GPT -6 Astra is officially here. It is now publicly available and only a few days after Anthropic released Fable 5 .1. So I spent all weekend testing these models with real work.
These are my actual workflows and other projects I've been working on, and I let each model run with them over the last 48 hours so I can see which one is truly the best model in the world for day -to -day work. the first task i sent both of them was to redo my website for some context this is what my existing website looks like so i specifically told them both that they could use this website as reference for information but not to copy the style i wanted each of them to come up with something original that hopefully would wow me so first let me show you what GPT -6 made inside of the ChatGPT desktop app.
So this is what we got right here. The first thing I noticed about this is that this looks a lot different from my original website, which I really appreciate because that's what I told GPT -6 to do. This has a lot more of a modern look to it with this more modular and boxy look.
I think it looks fine and it has a lot of the right elements in here. Like it has some videos that you can play. If you click on this, it pops it up.
into this viewer right here which looks really nice that's an improvement over my existing website where if you just click any of these it just plays it embedded on the page i think it did a good balance of not putting too much information in here while just putting in enough yeah overall it looks fine it's super clean it doesn't exactly wow me but it does work by contrast this is what fable 5 .1 made for us and immediately what i like about this website is that this feels more like my vibe i want my website to convey that i make content that's very friendly and approachable not super difficult for non -technical people and i think this nailed that vibe a lot more however i think it nailed it because it copied my existing website If you look at the two of these side by side, it is pretty clear that it copied some of the ideas from my original website, which is what I told it specifically not to do.
But I do think it looks pretty good. There are some elements on this page though that... aren't quite working so right here there is an arrow here that says that's me obviously that's in the wrong spot there are also these thumbnails that are floating around and as i've been playing with the page i don't know if i'm gonna be able to recreate it but every once in a while those get stuck in the wrong spot let me try scrolling all the way to the bottom and then to the top no but these would sometimes get stuck like up here so it's not exactly perfect and then we have these thumbnails right here for my videos and if i click on them It brings up this white box and nothing's in there.
And then if we scroll down further on this page, I think that Fable 5 .1 included too much information on here. It's a little bit too wordy for the type of landing page I'm going for. Even this section just has too much.
And I feel like it got a little bit lazy where it just kind of had these generic logos it created instead of bringing in the actual logos. for these brands that I've made videos about. And then down here, it went out on its own, which I thought was interesting, and found some of these resources I had published in the past and included them on my homepage.
That's not something I would have wanted. It is pretty neat that it found those on its own though and included them in here. So yes, overall, it's just a little bit too much, but I do like the look of it.
Oh, there we go. See, they just got stuck up there. That's what I was talking about.
In terms of functionality, GPT -6 won here. Everything on this site works. Every button I click does something, and clicking on the videos actually plays the videos in a pop -up.
And it followed my instructions and didn't copy from my existing website. But I do like the design from Fable 5 a bit better, even though it may have cheated. Next, I asked each of them to create a presentation for me.
Now, I'm not the type of person who likes to use traditional slide decks. Instead, I like to create these explainer websites where I can scroll through the website while I'm talking about the subject and using this as a visual guide while I'm recording a YouTube video.
This one in particular, I created last week using Fable 5 .1 inside of Claw Design. And I use this during my weekly AI news recap. And I was really happy with the way it came out.
which isn't really surprising because i'm always able to make really nice designs with clod design so you can take a quick look at what this looks like as i scroll through here we have these scroll animations we have this text and a lot of this is interactive so if we come down here to these benchmarks i can click on them and then you can see it appears like that so i'm really happy with the way this one came out so after gpt6 came out i ran the same prompt through gpt6 and this is what it created for me you can see it looks very similar because i gave it a lot of the same directions a lot of same design choices and guidance but you can see there it's not quite as nice because at the top here we have these dots in the background and then as we scroll down they disappear and just overall even though it looks very similar and it's a lot more interactive there are a lot more of these buttons to click a lot more interactive elements but it's not as smooth as what we have with claw design it doesn't quite feel as sharp when you use it this might be something that's hard to see in the screen recording that you're watching it maybe just be something that you feel as you use it
but it just doesn't have the smoothness that I experienced with the presentation website I created with Claw Design. But otherwise, they are pretty close in the way that they look. Now, I don't know about you, but I think that the presentations I created came out a lot better than the websites.
And that's because I gave the models more detailed instructions as to what I was looking for. And the designs were a little bit more straightforward than maybe a website. And that's pretty common with models these days.
You really need a human directing them to tell them what looks good and what they should create. Otherwise, what they produce on their own just comes out a little bit bland and tasteless. All that said, I think the winner for design work, even if it's just by a little bit, is Fable 5 .1.
Next, I wanted to see how each of these models were at vibe coding, specifically how easy it is for non -technical people to use these tools. for vibe coding so this first one admittedly is just for fun i asked each of them to create a playable game for me very similar to animal crossing if you're not familiar with animal crossing it's a game where you live on this island you do these little tasks like collect apples and fish and collect seashells then you can build up your home on your island it's a very chill game so let's look at the one that gpt6 made for us first so this is moss meadow And you can see this is my character right here.
I can move him around. I can go up to one of these trees to collect fallen wood. And here's another tree with some apples.
So I can collect these apples. I can talk to other characters like this one right here. And can go to this dock to do some fishing.
Here's another character I can talk to. And then I can come over here and sell all the stuff that I collected or buy some other stuff. And actually, in terms of functionality, I found that everything about this game works.
There are no game mechanics that are broken. And all this was one -shotted. I didn't tell it to update anything or fix any bugs.
This is exactly what it handed to me after it was finished for the first time. I also think it looks pretty cute. It kind of goes with the vibes of Animal Crossing.
Very cute looking. but it is a little bit small. The island that I'm on is not very big and there's not a lot to do, but effectively it is exactly what I asked for and everything works.
And then this is the one that Fable 5 .1 made for me. And immediately it looks a lot, lot different. In terms of looks, I don't like this one.
I think that the one that GPT -6 made looks a lot better. This one just feels a little bit cheap. But the mechanics are all great.
I can talk to characters like this. I can pick up things that have fallen from trees like apples. There are seashells that I can pick up.
Unlike in the one that was made with GPT -6, for this one I can actually go into my home and hopefully I can decorate it as well. I don't have any furniture to currently decorate it, but I could buy some furniture and then decorate my home. And so yeah, everything works.
I haven't found anything that's broken. So in terms of the mechanics, everything is fine. I just really don't like the way that it looks.
The game that GPT -6 made for me, I think looks a lot nicer. Next, I asked each model to build a real app for me. I asked for a to -do list app where I could just ramble into it and have it sort everything I'm saying into different task lists.
It's called Ramble Away. This is the one that was made by GPT -6, so let's give it a shot. All right, I got a lot on my mind right now.
I have to go to the grocery store later to pick up some ingredients for a salad, maybe some spinach and some onions. and some strawberries. And then I need to remember to take out the trash and walk the dogs before the end of the night.
So I said all of that. And hopefully that gets sorted into these different task lists. All right, pretty good.
It got the groceries right. It didn't repeat the strawberries, which is great. It added a new list called personal, a new one called home.
and a new one called work that's called take out so obviously that was a mistake but pretty good the basic functionality of it works i can delete any of these i can check them off i can edit them we can see the completed test here so yeah it's a pretty good start And here's what Fable 5 .1 created for us. And what I thought was pretty interesting the first time I looked at this was that with the website that we generated before and the presentation, I preferred the designs that were created by Fable 5 .1.
But for both the apps we created, the game and this app, I think that what GPT -6 created for us looks a lot nicer. But let's see how they actually perform. So we'll run a test with this one and I'll show you how it works.
When I go to the grocery store, I also need to pick up some bread and some milk. And then later today I have to call my brother and then tomorrow I need to make an appointment with the doctor. So what's cool about this one is that when you pause, it will treat that as one sentence and start transcribing it and sorting it right away.
That will speed things up a little bit. And then here it's properly sorting everything into family, groceries, and health. So both of these apps function very well.
It's exactly what I asked for. I just prefer the way that the app that was created by GPT -6 looks. There's another part of this we need to consider though.
For both of these vibe coding tasks, GPT -6 was a lot easier to work with as a non -technical user. With Ramble Away, for instance, it wrote all the code, then built the app itself, launched it. You can see right now it's using computer use to test out the app.
And then... When it was finished, it left the app open for me, ready to use. Fable, on the other hand, wanted me to run a script in Terminal and then to open the project in an app called Xcode to actually build it.
It was a similar experience when I asked them to build the games for me. Those games required me to have two apps on my computer, Blender and Godot. I originally asked Fable 5 .1 to install all those apps that it needed to build the game.
But what it ended up doing was asking me to run several commands in terminal and then to open up the apps myself to change some settings. So instead, I went over to GPT -6 and told it to install both of them, and it just did. I didn't have to change anything myself.
It opened up the apps and made all the settings changes it needed. And then it used them to create the games. I never even had to open the apps.
I actually have no idea how they even work. So for a non -technical user, if you want to vibe code anything, build some games, some apps, I think GPT -6 is much easier to use. Next, I asked each of the models to script out a YouTube video for me.
This is something I ask every model to do and every model fails at it because there's a lot of nuance to the way that humans speak and the way that humans write that AI really struggles with in terms of sounding human. So for the first test, I just told them to write me a script for a YouTube video that I can read out loud that tells people all about the new GPT -6 release.
And I purposely didn't give it the full context of my agent operating system. I didn't allow it to use any script writing skills or outlines that I've developed in the past. So starting with a completely blank slate.
And let me read for you some of what GPT -6 wrote. OpenAI just announced GPT -6. It's called Astra.
And the biggest question is whether it can finally handle the work we keep trying to hand over to AI. Because getting an impressive answer is one thing. Getting a finished result we can actually use, that's a much higher bar.
First, the rollout. So that sounds okay, but that... also sounds a lot like AI wrote it.
And if I was watching a video and that was the intro, I would probably click away because it doesn't sound super smooth or human -like, and I would know that an AI wrote that, and that would be a turnoff for me. I wanna hear from humans when I'm watching YouTube videos. So it's okay, maybe as a first draft, but definitely not.
a finished script that can just be read out loud so let's see what fable 5 .1 wrote so here's its intro gpt6 is out not a rumor not a leak openai shipped it on september 3rd and as of this weekend if you're on a paid chat gpt plan it's probably already sitting in your model picker and right away that is so bad that is so much worse for all the same reasons that sounds so much like what ai writes and i really don't like it So both of them fail that initial test.
But like I said, I ran this test twice because the second time I gave it more context. I allowed it to look at my prior transcripts from my prior YouTube videos and to get an idea of how I like to style my videos and my tone of voice to see if it could do a little bit better. So here's what GPT -6 wrote with that additional context.
OpenAI has released GPT -6 Astra. And the big promise is that it can handle more of an entire task for you from understanding what you want to checking the finished work. But that could mean researching a topic and turning it into a presentation or working through a project that involves several different apps.
So let's talk about what's new, how you can access it, and a few things you should know before you start using it. And immediately, I think that sounds much better. it's not perfect i would definitely make some changes to it but that is more usable than what we've seen but it's definitely not perfect because the more i read this the more i found that traditional ai slop type language here's an example of that right here it says and that second part matters even if you have absolutely no interest in coding these models love to tell us what matters and that language specifically where it says that second part matters is such an ai thing to say so it definitely doesn't pass this test but it is a bit better and here is what fable 5 .1 wrote openai just released gpt6 it's called gpt6 astra and it started rolling out on september 3rd and as of this weekend it's already showing up for plus pro and business users inside of chat gpt openai is calling it the world's most intelligent and aligned model that's a big claim so in this video i'm going to break down what's actually new
There's a little bit more, but I think that that sounds also much better. But then we get into the next section and it starts to fall apart. Quick orientation first.
It's been just over a year since GPT -5 came out. And in that year, OpenAI shipped a lot of point releases. Language like that, quick orientation first, reads AI slop to me.
And we have some more of it down here. It says, if you only remember one thing from this video, it's this. And then we have this.
Two things stand out in what they published. So just like with GPT -6, the results got better the more context I gave the model. But with both of them, you're still not going to get a perfect output.
You have to go in, you have to direct it, give it examples and polish up the final output to really make it sound like it's coming from you instead of from an AI. Comparing them side by side. I feel like GPT -6 did a little bit better here.
It sounded a little bit more natural and is more closer to the final product than what I got out of Fable 5 .1. Next, I asked each of them to edit a YouTube video for me. If you didn't know this, whenever I record videos for YouTube, I always have AI do the first pass when it comes to editing.
Kind of all the... boring stuff, removing all the silences, removing the false starts, and then it tries its best at adding the scene layouts and those zoom effects that I use. Now, usually I would use 5 .6 for this because it's been pretty good.
So I was really excited to try GPT -6. However, I was pretty disappointed at what GPT -6 produced. First, it took a very long time, and then the results...
aren't fantastic in fact i think it actually did worse than 5 .6 normally does it left in a lot of those false starts that it shouldn't have left in it didn't do a terrible job but it just didn't do as good as 5 .6 somehow what's really cool though and something i've never seen it do before is that on its own it blurred out sensitive information in the video you can see that right here there was a user key that was exposed and even though i didn't tell it to it was smart enough to blur that out.
That was super impressive to see. And then here are the results from Fable 5 .1. By the way, one reason I'm doing this test is because even though these use an MCP connection to do the editing, I also instructed them to use browser use to inspect the timeline to make sure it looks good.
So it's also testing computer use a little bit. Now, interestingly, Fable made... a lot of the same mistakes GPT -6 made.
It missed the same false starts and overused the zooms. It also did not blur out the sensitive info like GPT -6 did. And I think that's because the models had to use computer use for that.
And that's clearly somewhere that GPT -6 did better. So this is a win that I would give to GPT -6, but just slightly. The next benchmark was I asked both of them to execute a skill that I use every single day, which is my email triage skill.
This is telling the model to go through my email inbox and to decide what to do with all the emails in there and to draft replies without sending them. And I'm happy to say no surprise that both of them performed very well at this. But I would say that using GPT -6 or Fable 5 for something like emails is overkill.
You're going to get great results with weaker models like GPT -5 .6 or Claude Opus. This last prompt, though, is what really showed to me the difference between these two models when it comes to knowledge work. And this was the one, surprisingly, that really blew me away.
So this is the prompt that I gave it. I told both of them to act as an outside advisor, and I gave them access to a folder on my computer that I always use with my AI agents. It includes all the videos I'm working on, all the projects I'm working on.
everything inside my business so it had access to that file I'm telling you to go through that and just do a full audit tell me what's working what's not working where I'm losing money how I'm making money and what my next steps should be if I want my business to grow and anything in there that's personal what I can do to improve my personal life as well So over here on the left, we have ChatGPT and over here on the right, we have Claude.
And the way that GPT -6 approached this feels to me like the closest thing I have ever seen to AGI. Fable 5 .1 did a fine job, but it really approached this very narrowly and didn't ask me any questions as it was working. But GPT -6...
really considered a lot of outside factors that I didn't tell it about that were important for its analysis. And you can see that happening right here. It saw, for instance, that something that's really important for my business are my YouTube analytics.
And it correctly identified that inside my folder that I gave it access to, I didn't have updated reports from my YouTube studio with those analytics. So it asked me, hey, can you give me those?
I need those to make a proper conclusion here. Fable 5 .1 didn't do that. GPT -6 then correctly realized on its own that YouTube had changed the way that it counts views on videos about a week ago.
That is incredibly important for this analysis and GPT -6. figured that out and knew it was important enough to include in this analysis. Fable missed that entirely.
Fable 5 .1 also told me that one of the sponsors I'd worked with months ago never paid me because my bookkeeping records said they still owed me money. GPT -6 saw that as well, but then on its own, it went out, did some investigation by going into Gmail and realized I was actually paid and it was just the record keeping that was mismatched.
And so it seems like Fable was treating... everything in that folder as the source of truth without realizing that humans are bad at record keeping we forget to update things sometimes and so gpt 6 realized that it went out and checked on its own and realized that hey it was just a record -keeping mistake and i was actually paid and in general i'll say that the report i got back from gpt6 was a lot more useful than what i got from fable fable's breakdown felt like i was reviewing information i already knew gpt6 gave me real insights into my business that i was not aware of it was incredibly helpful and so it was this test and all the knowledge work i've been doing with gpt6 that has really shown for me the power of this new model it is really mind -blowing and i understand now why people are saying that this is close to agi this is close to a model that just understands a lot more and has an intelligence that we just haven't seen before.
It's really remarkable to work with. And I think that for knowledge work, this is where we're going to see the biggest difference. Overall, what's my take on this?
Well, let's be honest. If you use either one of these models, you're going to get great results. But a very important consideration is the cost from what i've been seeing you're going to get a lot more for the same price with gpt6 than you're going to get with fable 5 .1 and there have also been moments with gpt6 where i'm just like whoa it gets it in a way i've never seen a model get it before it's something you really have to experience for yourself and it makes me want to use gpt6 more just to see what else is capable of.
But that's just my take on it. Let me know your thoughts on these two models if you've been testing them out in the comments section down below. Otherwise, thanks so much for watching.
I really appreciate when you guys watch all the way to the end. It's the best thing you can do to help out my channel, so I do appreciate that. Thanks so much for watching, and I'll see you in the next one.
Bye for now.
The Hook
The bait, then the rug-pull.
A creator gave GPT-6 and Fable 5.1 the same real assignments, not benchmark prompts, and let each run for 48 hours across a website redesign, a presentation, two vibe-coded builds, a script, a video edit, an inbox, and a full audit of their own business files. The two models traded wins everywhere creative judgment was subjective. The gap that actually mattered showed up somewhere else entirely.
CTA Breakdown
How they asked for the click.
VERBAL ASK
25:00next-video
“Let me know your thoughts on these two models if you've been testing them out in the comments section down below.”
Soft, low-pressure comment-bait closer rather than a product pitch, tied to genuine curiosity about viewer experience.
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 launch-day test of Anthropic's new default Opus model — benchmarks, a landing page build, a script draft, and a motion graphic, all run head-to-head against Fable 5.
OpenAI, Anthropic, and Google all released flagship models in the same week, Gemini Notebook quietly capped how much you can generate, and Grok's new payments plugin lets a bot spend your money with your sign-off.
A YouTuber walks through the eleven-bot staff living inside Grok Bot, each with a name, a job, and a cloud computer that keeps working after he closes his laptop.
A YouTube tutorial on wiring ChatGPT and Claude into a $79 desktop editor called Borumi so they cut, zoom, and animate your talking-head videos for you.
00:22
00:49
00:59
01:17
01:38
02:03
02:48
03:01
03:18
03:46
03:51
04:08
04:34
05:13
05:31
05:50
06:09
06:28
06:52
07:06
07:26
07:43
08:21
08:33
08:59
09:18
09:44
09:56
10:34
10:45
11:12
11:31
11:52
12:15
12:28
12:46
13:05
13:18
14:04
14:31
14:40
14:59
15:18
15:42
15:56
16:34
16:48
17:12
17:31
17:49
18:10
18:27
18:46
18:59
19:24
19:43
20:21
20:33
20:59
21:18
21:37
22:15
22:30
22:52
23:11
23:30
23:49
24:08
24:27
25:05
13:02
07:18
24:02
15:58
13:35
12:18