Full transcript
0:00I just put GPT6 Astra head-to-head
0:03against Fable 5.1 on seven realistic
0:06business use cases to see which one is
0:08better for real work. I'm talking things
0:10like turning Loom videos into
0:12documentation, creating launch videos,
0:15building landing pages, generating ad
0:17creative and computer use. For each one,
0:20I'll show you which model did better,
0:21how long it took, and most importantly,
0:24what it cost me. By the end of this
0:26video, you'll know exactly which model
0:27to use for each job and which I'd keep
0:30if I could only have one. The problem is
0:32that most comparison videos out there
0:34right now are giving you unrealistic
0:37expectations of how these models perform
0:39in real world conditions. So, in this
0:41video, I'm testing each model in those
0:43real world conditions. I set thinking to
0:45high. I'm using prompts that a normal
0:47user would actually give, and I'll be
0:50running both through their respective
0:52desktop apps. Each model will get
0:54exactly the same prompts and the same
0:56context. And to make this as fair as
0:58possible, neither model can see the
1:00outputs of the other. I won't intervene
1:02except to answer any clarifying
1:04questions the model asks me or to stop
1:06it if it's going totally off track. The
1:09token costs you'll see me go through are
1:10API priced estimates for work covered by
1:13my normal subscription. Some of the
1:15tests incorporate extra tools like
1:17Higsfield MCP and paper.design. These
1:20tools are not included in the cost
1:22estimates. All right, so without further
1:24ado, let's get into it. So the first use
1:27case is Loom documentation, which is one
1:29I'm doing heaps myself. So essentially
1:31what I will do is record a Loom video
1:34walking through something, whether that
1:36is a workflow that I've built, an agent,
1:39or literally anything that I want to
1:41talk to the client about. From there,
1:42I'll take that Loom video and then
1:44convert it into a HTML document, which
1:47will have a step-by-step walkthrough of
1:49exactly what they need to do. This is
1:51the Loom video that I recorded.
1:52Essentially, it's just a 2-minute video
1:55that steps through the process of
1:56installing the watch skill from GitHub.
1:58So, super basic. What I asked Fable and
2:00Astra to do was to take that Loom video
2:03and then turn it into an exceptional
2:04client-f facing HTML guide. I wanted it
2:07to watch the full recording, extract the
2:09relevant screenshots and instructions,
2:11and then make it a swipe through
2:12step-by-step experience so that when
2:14someone opens the site, they only see
2:16exactly what they need to do in that
2:18step, and it avoids a lot of the
2:19overwhelm. Payable actually worked for
2:2118 minutes on this workflow. Here's what
2:23it gave us. So, we've got the
2:25step-by-step section. We've got some
2:27nice transitions going between each of
2:29the sections. On each of the
2:30screenshots, it's added that orange box,
2:32and the orange box is sort of pulsing,
2:34so it draws your attention, tells you
2:36exactly where to find things. And then
2:37when I hover over the actual
2:38screenshots, it has this nice magnifying
2:40effect. I can click through step by
2:42step, and it shows me exactly what I
2:44need to do for each step. And the
2:46transitions between each of the pages
2:47are really nice. And I actually really
2:49like how simple it is because I can just
2:51look at this page and I don't feel
2:52really overwhelmed at all. Then I can go
2:54up to full guide and see every step one
2:56by one. And if I wanted to, I guess I
2:59could print this out. And it looks like
3:00it prints pretty decently. It doesn't
3:02look like Fable's actually optimized for
3:03that, and I didn't ask it to, but that's
3:04something that customers usually do is
3:06want to print these. So, let's compare
3:09that to what Astra did. So, Astra has
3:11done a completely different design. Um,
3:13and what I immediately notice when I
3:15open this up is that it feels cluttered
3:18and a little bit confusing just cuz
3:20there's so much text on screen. What I'm
3:22just worried about is these screenshots
3:23mainly are a little bit cropped. So,
3:25it's hard to follow where exactly I need
3:28to be looking. And these highlights are
3:30actually a lot better than they are on
3:32Fable in terms of accuracy of the
3:34highlight. But the highlight themselves
3:36sort of blend in because everything's
3:37green and it doesn't like pulsate at me.
3:39And then nothing zooms in when I hover
3:41over it. Especially considering how
3:43small these screenshots are. Uh I would
3:46probably want that magnification
3:48feature. But when I do go to enlarge, it
3:49does enlarge. But that enlarge mechanism
3:53is like a little bit less nice than what
3:55Fable had. So again, I think Fable Fable
3:59here has a more polished design for this
4:02essentially as far as like steps and
4:03functionality. Both of them are exactly
4:05the same. They've captured the same
4:07steps and those steps are both accurate.
4:09Astra has actually given us that print
4:11and save as PDF button though, which is
4:13really, really nice and something that
4:15Fable actually missed out on despite
4:17taking over double the amount of time to
4:18do. So, taking a look at the cost
4:21breakdown for this, um, Fable costs
4:24$9.89
4:26and Astra cost $41. So, they're about
4:29half. And that goes for time as well.
4:32Fable took 18 minutes. Astra took about
4:349 minutes. Guess the question is, you
4:36know, do is it twice as good of a
4:38result? Probably not twice as good of a
4:39result. But if we're analyzing this
4:41purely on output alone, I'm definitely
4:43going to pick Fable here as the winner
4:47because that output was just so much
4:48easier to understand. Each of the slides
4:51made more sense. Um, I wasn't feeling as
4:53overwhelmed as I was when I was looking
4:54at the chat GVT one, which I think just
4:58because it had so much text on the page
4:59and if you gave that to a client, they'd
5:01probably be a little bit, you know,
5:02confused. Next up is video editing,
5:04which is something I'm really excited
5:05about cuz I've been working on this tons
5:06on the channel recently. So, I've given
5:08both models this prompt, which asks it
5:10to create a 30-cond promo video for a
5:15cohort that I'm thinking about launching
5:16in November. Essentially, just building
5:18up some hype for it and getting people
5:20to join the weight list. So, what I've
5:22asked it to do is read some of the
5:24launch plan that I've written down
5:25inside of my OS and then from there
5:27choose some music, some SFX, some
5:29existing audio files from my library
5:31that I have inside of my existing video
5:33editing project. So, what I'm asking to
5:35do is tell a clear story about who this
5:37is for, what we'll build, and why
5:38someone should join the weight list. I
5:40wanted to build some suspense and make
5:41the value clear through graphics and
5:43text alone. Not asking it to do an AI
5:45voice over or anything like that. So,
5:46both these models got the same one.
5:48Let's see what Fable did. So,
5:50immediately, I really like that first
5:51section here. It sort of has the
5:53clicking effect [music] and it's as if
5:55I'm typing on a terminal. Um, the
5:57background is kind of cool. Um, we've
5:59got some nice clean transitions in
6:01between everything. I really like this
6:04section here with the text coming in
6:05sort of like as if you're typing into
6:07chord or or using the terminal and sort
6:10of highlighting those pain points. And
6:12then eventually what I did like was just
6:14how well it timed it to the the music
6:16that it selected. So this section here,
6:19so like that founder OS title sequence
6:21comes in at the perfect time there and
6:23right on the music cube. So that was
6:25Fable's attempt. Let's have a look at
6:27Astro's attempt now. exactly the same
6:29prompt. And let's check out what it got
6:32up to. So, Astra's attempt is really
6:35off-brand. Um, just out of the gate. Um,
6:38it's invented a new color scheme, which
6:40is fine. Like, I didn't specifically
6:41instruct it to go out and look for my
6:44color scheme or go and find anything.
6:46Both both options were off-brand, and I
6:48think both of them have mistakes as far
6:50as exactly what's going to be in the
6:52cohort and and the [music] offer itself.
6:54But, that's okay. I think that Astros
6:56has a little bit less character to it
6:58and a little bit less personality. The
7:00other thing I notic is that the music
7:01itself isn't timed [music] well to
7:04uh to the assets or the motion graphics
7:06on screen. We have a look at the
7:07costing. I mean, there's a massive
7:09difference in cost and time. So, Fable
7:12took 38 minutes and 16 seconds to come
7:15up with that video and cost $20 compared
7:18to $6 and just under 10 minutes for
7:21Astra. So, I mean, you can see where the
7:24difference has gone. I think Fable spent
7:26a lot more time going over it and timing
7:28everything, which sort of shows in the
7:29final result. To be honest, I think I
7:31could probably get away with using that
7:33first output that Fable gave us. I
7:35definitely couldn't use the one that
7:36Astra gave us. So, again, this is
7:38another win for Claude. Use case three
7:41is building a SAS landing page. So, what
7:44I've asked both models to do is build
7:45the most impressive SAS landing page
7:47that it can. It's going to choose any
7:49SAS concept that you like that you'll
7:51think will work best. Invent its brand
7:53positioning and copy. Research
7:54outstanding landing pages online and
7:56identify what makes them exceptional.
7:58Use that research as inspiration for an
8:00original design. And let's see what
8:03Fable has come up with. So immediately
8:06we've got some nice transitions coming
8:08up on the right hand side. Got those
8:09colors. I'm not sure what this product
8:11is. We've got some really distinct
8:13claudisms I guess you could call it for
8:15like the design elements that are coming
8:17in. Um, so it's very obviously AI
8:19generated. Uh, the the transitions
8:22though in motion itself is quite smooth.
8:24So my so my immediate reaction is that
8:26this is a pretty meh website from Fable
8:28here. Um, nothing that I would sort of
8:31consider as like super out there. So
8:33let's have a look and see what Astra
8:34came up with. Immediately it's a much
8:37nicer looking hero section. Um, I think
8:40at least. I don't know how it's created
8:43this
8:45element on the right hand side, but
8:46you've got some nice I guess you' call
8:48it parallexics. You got some nice
8:50animation there as it is shifts down
8:53while you scroll. I think it's a cleaner
8:56sort of setup. As we scroll down, we've
8:58got some sort of nice sections here.
8:59This is a cool little animation of a
9:01calendar. Starts off crooked and I guess
9:04it, you know, eliminates wasted time and
9:06adds some focus time in your calendar.
9:08It makes it look a little bit better.
9:09Got an ROI calculator down here and it's
9:13dynamic. Everything sort of moves around
9:15and Liz. So for the SAS landing page
9:18that cost Fable about $14 in tokens and
9:22it took it 34 and it took it 34 minutes
9:24to complete that website compared to
9:27Astra which took about 23 minutes and
9:30cost uh a little bit less $13. Very very
9:33clear winner for this one for me. That's
9:35Astra. I mean, Astra's website here is
9:38significantly less immediately AI than
9:42Fables. Fables screams AI SASS landing
9:45page, and you've probably seen thousands
9:47of these. Um, Astra here does have a few
9:50AISMs or or Chachibisms with like the
9:53little sections and the numbers coming
9:55up, but that's not something that I'm
9:57actively prompting against in these
9:58tests. I'm just trying to see what the
10:00models will do with a a basic prompt
10:02that most people are going to give and
10:04trying to see which one gets the best
10:05results. So obviously you can make far
10:07better websites than these results with
10:09AI. So at the moment we're three points
10:13in and it is 21 fable. So use case so
10:17use case number four is ad creative. And
10:19by the way this video is sponsored by
10:21Higsfield. Higsfield gives you access to
10:23image and video generation models in one
10:25place. So you can create things like
10:27product photos, add creatives, and video
10:30clips all with the all with the MCP
10:32connector, which lets Astra and Fable
10:35connect directly into these models. Now
10:37I can ask it to invent an energy drink,
10:39a brand, generate the product images
10:41through Higsfield, and use all of those
10:43images to create three ads. Astro works
10:46out the brand, the copy, and what each
10:47image should do, and then it talks to
10:49Higsfield to get those images to
10:50generate. If you want to try Higsfield,
10:52I'll leave the link in the description
10:53below. For this comparison, both models
10:55get the same creative process and use
10:57the same image model through Higsfield.
10:59They still have to invent the product,
11:01pick the buyer, and come up with the
11:03campaign themselves. So, let's see what
11:04they did. So, Fable decided to come up
11:06with some with a brand called Small
11:08Hours, an energy drink for people who
11:10work the night shift. Uh, and it's these
11:13are the ads that it generated. Um, so
11:16this is the first one here. The 4 a.m.
11:18wall doesn't the 4 a.m. wall doesn't
11:20care that it's your third night. I mean,
11:22the copy there is a little bit difficult
11:24to read. I had a little trouble. Um, the
11:27image itself is pretty good. The
11:29branding on the bottle I'm not 100% sure
11:31on. 3:00 a.m. is not the time for 200
11:33mg. Yeah. Try the first pack. Some
11:36pretty decent ads there. This one looks
11:39a little off to be honest as far as the
11:40the car prompting for the shift that
11:42ends when everyone else's start. And
11:44then you've got the actual product image
11:45itself, which looks pretty pretty cool.
11:48Um, it's got like a matte can finish,
11:51all black, small hours, and the sunrise
11:53coming through. So, that's Fable's
11:55attempt at this. Now, let's have a look
11:57at Astro's output, which is way, way
12:00different. But, let's start with the
12:01product photo here for what Astro has
12:03decided to call Odd Hour, a pink
12:06grapefruit, and rosemary uh energy
12:08drink. So, two very different target
12:10markets here. So, Fable's going after
12:12night shift workers, like you think
12:13nurses, doctors on their way home from
12:15night shift. Uh, Astra is going after
12:18creatives um, who at 3 p.m. just need a
12:20little hit to keep going for the rest of
12:22the day. So, this is the product image
12:24that is generated. I think they're more
12:25creative. Like, make 3pm weird is much
12:27more concise copy. Okay, less sweet,
12:29more bite. I like the concept here, but
12:32you've got some really bad GPT image
12:35texturing going on on the grapefruit,
12:37which I think, you know, I think if
12:39you've used GPT image 2, that is an
12:42issue. It doesn't look like it's been
12:43fully resolved here with 2.5 either, but
12:47to be honest, these are significantly
12:49better ads. The they're really, really
12:51nice ads. So, let's have a look at the
12:53cost time and tool calls. So, for ad
12:56creative, it cost $49 for Fable to
12:59generate those image and took 8 minutes
13:02and 54 seconds as opposed to Chacht,
13:04which cost $5.18
13:07and took 11 minutes. And this is
13:09actually kind of surprising to me
13:10because Fable up until now has been
13:12taking significantly longer to do
13:14anything compared to ChachiBT. I think
13:16we've got another win for Astra, which
13:18means that we're all tied up two and
13:20two. Um, but in in this comparison,
13:23let's look at the next use case, which
13:24is ebooks. So, so for the ebooks, I'm
13:26going to be using a tool called
13:27paper.design, design, which is a
13:29decently new tool, which is sort of like
13:30AI native Figma, and it's something that
13:33I've been using a lot recently for some
13:34of my creative workflows, and it's
13:37really, really good at creating lead
13:39magnets like PDFs, ebooks. So, I want to
13:41put paper and these new models together
13:44and see which one can get the best
13:46results. So, this is the prompt that I
13:48have given both models. Essentially, the
13:50task is to create a GPT6 Astra versus
13:53Fable 5.1 ebook lead magnet in paper.
13:56This is going to be the lead magnet that
13:58I'm actually putting in the description
13:59below, which has all of the prompts that
14:01I'm giving out for today and all of the
14:02detailed breakdowns of the results. So,
14:05the first thing that Fable did is that
14:06it actually tried to go out and grab a
14:09bunch of the visual designs from the
14:11websites, but it wasn't able to make
14:13those fit with the design that it wanted
14:16to do inside of paper. So, it ended up
14:18using Higsfield and Chat GBT image 2.5
14:21to generate images uh that it could use
14:24as part of the paper uh design as part
14:27of the ebook that it wanted to generate.
14:29So, it went through and it generated a
14:31bunch of different ones. It read them
14:33all and added them directly. And if
14:35you've been following the announcements
14:36here, you these are going to be super
14:37familiar. We've got the Galaxy that sort
14:39of accompanied a lot of the marketing
14:41for GBD6 Astra and then Fable which had
14:44more of like a blue sky pink clouds sort
14:46of vibe to its marketing. Now
14:47immediately this cover page looks pretty
14:49good. Um we've got Fable on the left,
14:51Astra on the right. We've got both of
14:52those AI generated images. Um this is a
14:55little hard to read having the text
14:57across both different backgrounds. So
14:59that's probably not the best. Nice
15:01little um drop shadow there. Each of the
15:04pages is actually looks like it's gone
15:05and tapped into my existing brand
15:07guidelines, which I didn't ask it to do.
15:09I asked it to just generate um generate
15:12some images. We've got all of the
15:13different prompts and we've got all of
15:16the different breakdowns of the results.
15:18So, as far as this lead magnet goes, I
15:21think it's okay. Like, uh it's got it's
15:23got everything that I needed to do. It's
15:25definitely getting giving some more of
15:27those AISMs or claudisms specifically.
15:30So, let's have a look now at what Astra
15:32did. So, Astra got the exact same prompt
15:34and it also had Higsfield and the paper
15:36MCPs connected exactly the same. So, the
15:39first thing that it actually did was ask
15:40me a question, which I think Astra does
15:43a lot more than Fable. Astra seems to
15:46want to confirm your intent a lot more.
15:48And that comes back to some of the stuff
15:49I've been reading that Astra itself is
15:52better at capturing and understanding
15:53your intent than some than Fable is. It
15:55did ask me a question about the link to
15:57the landing page, which hasn't been
15:58published yet. though doesn't exist for
16:00the CTA at the end of the PDF. So I just
16:03asked it to include an empty link rather
16:06than you know create one. So that's a
16:08good pickup that Fable itself didn't
16:10actually capture. Then it's gone and
16:13used Higsfield again. And the
16:14interesting thing here though is that I
16:17think Chachbt didn't even bother going
16:19out and trying to use those external
16:21resources that are found inside of the
16:23design. Instead, it took those as
16:25inspiration, passed them as references
16:26to Hicksfield, and then generated the
16:28new concepts. I think that Astro's
16:30design here is much much nicer. Now, a
16:32couple reasons why I think that I think
16:34that the placement of the galaxy that it
16:37uh generated is a lot nicer where it's
16:39over on the left hand side and sort of
16:41complements this half and half look that
16:43it's gone for. None of the text is
16:45overlapping that sort of cut off so it's
16:47easy to read. And then as far as the
16:49layout, everything here looks a lot more
16:52professional. So I think the problem
16:54with Fable's execution is that its brand
16:56guideline, the execution of my brand
16:58guidelines wasn't very good. Uh Astra
17:01has ignored the brand guidelines totally
17:04and just gone for something that's going
17:06to fit the design of the cover page and
17:09these two uh new models that are coming
17:12out. So to be honest, I prefer Astra's
17:14output here. Me from the future here. I
17:16was so wrapped up in Astra's ebook that
17:19I completely forgot to go over the
17:20stats. Astra took about 21 minutes and
17:23$16 in tokens compared to $26 and $17
17:28for Fable. So, it gave me a cleaner
17:30guide in less time for basically the
17:33same cost. Astra is the clear winner,
17:35which puts it in the lead at 3:2. Let's
17:38take a look at the second last use case
17:40now, which is using Astra and Fable to
17:43control my computer and do things that I
17:45can't typically do with MCPS or APIs.
17:48So, I use Kit for my newsletter and lead
17:50magnets. And before every single video
17:52that I post, I have to come in here and
17:54create a new landing page so that you
17:56guys can come in and grab the PDFs and
17:58resources that I make for you guys and
18:00also then create some workflows inside
18:02of Kit. So, here's the prompt that I
18:03gave to both of these models. I wanted
18:05to use computer use to jump into kit and
18:07duplicate a recently created landing
18:09page. From there, customize it to this
18:12particular video, go and create the
18:14workflow tools, go into the workflow
18:16editor and create a workflow that'll
18:18assign the new tags that it creates to
18:20this workflow. And then I wanted to test
18:22the entire thing in my Gmail to make
18:24sure that it all works and it's ready to
18:26publish. So, let's take a look at how
18:27Fable did in five times speed. You can
18:29see that it's just as soon as I give the
18:31prompt, it starts to think for a while.
18:33Let's skip through that because Fable
18:34takes a really, really long time to
18:37think. Then it starts up computer use.
18:39So, it has jumped straight into my
18:41Google Chrome and you can see the orange
18:44cursor there. That's Claude. It's
18:45duplicating the book to skill guide.
18:47It's doing the confirmation email,
18:49changing everything there, and then it's
18:50handling the book to skill upload. So,
18:53essentially, that's taking Claude about
18:55an extra 5 minutes just to tidy
18:57everything up there. But, it handled
18:59everything really, really well. Now,
19:01let's take a look at how Astra did. Now,
19:03Astra took a little bit longer to
19:04actually get this done. So, let's watch
19:06it on five times speed again. So, Astra
19:08went straight into Kit again and has
19:10started to duplicate the original book
19:11to skill guide using the exact same one.
19:14Went through and it renamed it
19:15correctly. Astra did get a little bit
19:17confused here when it's renaming the
19:19landing page. It's adding things that it
19:21shouldn't be adding and going through. I
19:23actually had to intervene with Astra two
19:25or three times here um because it was
19:27just overly cautious. So the PDF output
19:30that it was exporting from paper
19:32couldn't be saved into the output folder
19:34that I had given Astra to work in for
19:36these tests. So it was asking me for
19:40permission to actually go and move them.
19:41Now it's building out that workflow and
19:44it started off really strong here. It's
19:45confirming the PDF creates the correct
19:47actions, adding the right tags, creating
19:49everything perfectly. So it starts to
19:52veer off a little bit here. It starts to
19:53add a condition. again being overly
19:55cautious trying to add a filter that
19:58would prevent any prevent this workflow
20:01firing for anyone else apart from my
20:03testing email. Uh it got I actually
20:06intervened here again just to stop it
20:07from doing that and delete that filter.
20:09So it came in and got rid of it. I
20:11actually had to give it more
20:12instructions as to how to find that
20:13embedded sharing link as well. So
20:16overall, Astra here is a little bit
20:19worse than Fable, which is super
20:21surprising for me because a lot of the
20:23press that's been coming out of OpenAI
20:25is how much better Astra is supposed to
20:28be at computer use. And this is
20:29something that I've been seeing from a
20:31lot of creators, but in this specific
20:32test, it didn't really hold up. Uh Fable
20:35really overperformed here. So for
20:37workflow building, let's have a look at
20:39the breakdown. For Fable 5, that cost
20:42$10 to do, took about 15 minutes. Astra
20:45took $24 over twice as much and about
20:48the exact same time. I think that comes
20:50down to just significantly more tool
20:52calls. You got about another hund
20:54another 50 tool calls than what Fable
20:56did. Fable will sit there and think and
20:58think and think about the action it
20:59takes before it takes it. Spending more
21:01tokens, taking more time most of the
21:03time. Astra is more likely to jump in,
21:06think less. It feels faster when you're
21:08using it. So that might be what you're
21:09looking for. But at the same time, a lot
21:12of the time when it is faster, it's
21:13actually getting things wrong. So, for
21:15this one, I'm going to give Fable the
21:17win here, which to me was incredibly
21:19surprising cuz I was very much expecting
21:21uh Astra to come out on top on this one.
21:24Now, the last use case here is a sales
21:27presentation or a slide deck or a
21:28PowerPoint or whatever you want to call
21:30it. Essentially, I want Fable and Astra
21:32to go headto-head building out the
21:33prettiest, best slide deck that possibly
21:36can be made for sales. So, here's the
21:38task that I gave both of these models.
21:39Create an original pitch deck for my AI
21:42automation services. You may read
21:43factual context about me and my
21:45business. Develop the story, slide
21:47structure, and design yourself from
21:48scratch. So, I don't want it bringing in
21:50anything from my existing design
21:52language for this test. So, this is the
21:53slide deck that Fable came up with. And
21:55you can see first that immediately I'm
21:57pretty impressed with the design here.
21:59It doesn't [clears throat] feel
22:00extremely AI. Obviously, we've got some
22:02AISMs creeping in, but I'm I'm liking
22:04how it's structured. Some of the spacing
22:06is off here. Again, some spacing issues.
22:09Some graphs here aren't perfect. This is
22:11a nice one. I actually really like that.
22:13Um,
22:16some good graphs. Very professional
22:19looking sales pitch. Got the investment
22:22page, ROI figures, and then start with
22:24the solution order. Generally, like
22:26obviously not on brand because I
22:27specifically asked it to not be on brand
22:29and create something from scratch. And
22:31honestly, I'm really liking how what
22:34Fable's done here. There is a lot of
22:36really nice diagrams. I think this slide
22:38deck itself is actually very usable in
22:41its current state. There are a few
22:42things that you'd probably want to clean
22:44up and you probably wouldn't put this in
22:45front of the client straight up. So,
22:47let's compare that to how Astro went.
22:49All right, so make room for better work.
22:52Immediately, I like the headline better
22:53a lot on this one because just a lot
22:55clearer. I find Astro's copyrightiting
22:57to be a little bit better than Fables
22:59recently just cuz it's able to string
23:01together sentences in a more coherent
23:03way than Fable can. Sometimes Fable and
23:05Opus and all of the Claude models have
23:07this tendency to sort of overexlain uh
23:10go over the same ground again and again
23:11and again and it comes to this point
23:13where you're like what am I actually
23:14reading now? Generally like a nice
23:17looking slide. I would say less
23:18professional than the last deck I've
23:21seen um from Fable here and some minor
23:24formatting issues. Yeah, look, it
23:26started strong with the copyrightiting,
23:27but I think this deck is a lot weaker
23:30overall than what we're seeing from
23:33Fable here. I think that it's just a
23:35little light on content. Um, it's
23:37definitely designed for to more speak to
23:40uh and have less notes than the previous
23:42one has a little more more detail and it
23:44resembles more of the pitch decks that
23:47I've seen out in practice previously. So
23:49before we crown our winner for the sales
23:51presentation, let's take a look at the
23:53pricing between these two options.
23:55Claude took Claude cost $6.15. Astra
23:59took $244 and 6 minutes uh to create
24:03that slide deck. Again, Fable here has
24:05taken twice as long, but it has given
24:07us, I would say, a twice as good result.
24:10So again, we're seeing the exact same
24:11trend that we've seen for all of these
24:12use cases where Astra is much quicker.
24:14It can get you a decent result fast and
24:16cheap. Fable takes a very long time, but
24:19that result tends to be a little bit
24:21better. So, the winner for use case
24:23seven, I'm going to say, is Fable 5. So,
24:25after all seven use cases, the final
24:27score here is 43 to Fable. Now, that is
24:30an incredibly close result, especially
24:32considering how much hype that Astra is
24:34getting. Now, a lot of these were
24:36really, really close. I think there's a
24:38I think there's an argument, especially
24:39on number six, workflow building, that
24:41that should be an Astra win. These
24:44models are very, very capable. Both of
24:46them are extremely capable. You could
24:48use both of them to have success. I
24:50think what this test sort of reflects is
24:52that they behave in slightly different
24:54ways. You will see Fable think longer
24:57and take more time to get to its first
24:59result than Astra will. Astra will give
25:02you a much faster result. You it'll feel
25:04snappier, but you'll have to go back and
25:06forth with it. I guess that's the
25:07trade-off that you really have to make
25:09with it. Essentially, a lot of the time
25:12you're probably going to spend that
25:13money with Astra anyway, but you're just
25:16going back and forth with it. I mean,
25:17there are some cases here where we're
25:19seeing significant departures, like the
25:22promo video where, you know, Fable took
25:2438 minutes and cost over two and a half
25:27times more uh compared to 9 minutes for
25:30Astra. Now, was the result from Astra
25:32and Fable pretty comparable there? Yes.
25:34So, it's a very strong result from
25:36Astra. Fable's extra time and money
25:38there did result in something that we
25:39could actually use. Astros we couldn't
25:41without significantly more revisions and
25:43edits. And then who's to say that at
25:45that point that you're not up close to
25:46the amount of time that it's taken Fable
25:48to do that first pass. If I had to
25:50summarize the difference between Fable
25:52and Astra, it's this. Fable is a much
25:54more thorough model. It spends more time
25:57thinking about its next steps and
25:58testing once it's actually made a
26:00change. That means it's more expensive.
26:02It means it's slower. But nine times out
26:05of 10, you'll get a better result from
26:06it the first time. Astra is less
26:08thorough, but it gets you 80% of the way
26:10there in much less time and much
26:12cheaper. And in some areas, it
26:14absolutely runs laps around Fable. On
26:16some of the design tests, after some
26:18thinking on it, I think it's possible
26:20that I've just seen so many claw designs
26:22that I am better at picking up those
26:24patterns than Astros. And since GPT has
26:26been so bad at design in the past,
26:28seeing something new looks really good
26:30to me right now. Now, if I had to choose
26:32just one of these models to use for the
26:34rest of my life and usage limits were
26:36not a factor, I would choose Fable. But
26:39if I was on budget, I'd pick Astra. And
26:41that and that comes down not just to the
26:42results, but the way that the plans
26:44work. On my current OpenAI plan, I can
26:46put my entire weekly limit towards
26:48Astra. With Claude, I can only use about
26:50half of my weekly limit on Fable. So,
26:53that makes Astra pretty compelling if
26:55you're trying to get the most Frontier
26:56Intelligence possible on one
26:58subscription. Want to leave you with one
27:00final thought on this. The model isn't
27:02the bottleneck anymore. These models are
27:04great, but so is GPT 5.6 Soul and Fable
27:075. And the way that you win with AI
27:09today isn't by obsessing over which
27:11model is best. It's by building out your
27:13own setup that takes real work off of
27:15your plate. If you want to learn how to
27:17do that and 10x your productivity in the
27:19process, check out the weight list
27:20below. We'll take AI and apply it
27:23directly to your business with a setup
27:24you can run from one folder. You'll also
27:27find the free guide there with all seven
27:29prompts and the results from this video.
27:31This was useful, subscribe. Thanks for
27:33watching.