Gpt-6 Astra vs Fable 5.1... 7 Real Life Use Cases

0:00 I've been testing chat Astra and Fable 5.1 since they dropped. And in this video, I'm going to compare them on real life tasks to show you which is better. We're going to look at cost. We're going

0:09 to look at speed and my ultimate recommendation as well as the thing that most people are completely missing out. Now, we're going to do this across seven different levels. So, if you haven't

0:19 grabbed that coffee, make sure you've got it. Make sure it's beautiful. And we're going to dive straight in here. Now, this is the document that's going to go through basically all the results.

0:26 I'm going to be touching on the tokens and who won that in a second, but let's begin with the websites. So, let's go ahead and open these two up in a brand new tab. And I'm also going to touch on

0:35 some things that um you're going to find very helpful. So, this is the one that chat GPT Astra came back with. As you can see, I've got this kind of like scrolling wheel. And bear in mind, I

0:44 didn't do any special prompts. I basically said, "Hey, go and build me a website, bro." And this is what it came back and did. So, build things you didn't think you could explore the

0:55 experiments. It's got it's basically pulled in a lot of assets. I do really like this border that it's got here. I think that's quite nice. I can pause the motion. I can play it. I can do anything

1:04 I want to less theory, more look what I made. That positioning I probably would change a little bit, but you can see I got some of my individual things here. I think

1:14 it's got a good animations to hover over it, which is cool. Again, I've given it no. I've done no no skills is just going raw. And I like what they've got here. This feels fresh, quite new. Next idea

1:23 deserves a first try. Jack Roberts at the bar comes. This is what it just one-shotted, went ahead and did something cool for me. Then on the other hand, we have here Fable. Fable's

1:32 slightly more busy. I like the star, but it's taken how it's got two different carousels running up there. Now, what's cool here is built with AI on screen for real and then what do you want to build

1:42 today? So then I can come down answer some questions. Let's go down and click on continue. And then maybe based on that I it's actually organized it into different categories. I actually think

1:50 that is a really beautiful touch to be completely honest with you and it's got some information about me start building. I think from a functionality perspective I'm going to give the first

2:01 one to fable 5.1 just for it kind of like understanding and thinking about well what might they want to do on the screen which is cool. I would say from a design point of view

2:10 though that chatbt Astra has taken it. Now bear in mind the thing to understand is like you know if you get design chatbt built this website on a video I did previously okay and it did it on a

2:23 oneshot. This is when we're using the skills that pay the bills but in order to make this test fair I wanted to strip away the skills and just make it something that's really like go ahead

2:33 and do it. So this is what trackbt Astra is capable of. These models are freaking incredible and you've seen all the amazing websites that were built with Fable 5.1. Awesome. But just from a

2:42 website point of view, both are doing very well. Now, typically speaking, when you're going to be building a website with these models, you want to give it incredible references. I'll put a link

2:50 down below for basically like a gazillion these references that you can just grab and play around with. So, I think in this one, I'd say it's fairly even. Both are good at designing, but

3:00 I'd say I'd give the edge to Astra on this one with a capability edge. Probably going to claw. But again with one prompt a little more specific we could have gone either way. That takes

3:10 us on to the second one which is going to be copy and strategy. So essentially the challenge I gave them here was look I want you to get Sam Olman to say yes to a 30inut conversation with me. And by

3:25 the way if you're new here I'm Jack. I built and sold my last tech startup with like a gazillion customers. Now I'm building my own a startup and I share stuff on this channel uh that actually

3:32 works. So that's what we cover here. If you like that, feel free to let me know down below. Uh, drop us a like and a subscribe. It's appreciated. Now, essentially, they're going to do a cold

3:40 email for a 30inut Sam Alman interview. Uh, it's supposed to be five crisp strategic moves tailored to me. So, let's go ahead and check out what they've actually physically done. So,

3:49 why don't we go ahead and begin with Astra here. What would you let AI finish without checking? Hi, Sim Roberts. I teach practical AI workflows on YouTube from computer use to building business

3:57 systems. I'm putting two AI systems through seven real jobs, including website, video insert. Would you join me for a 30-minute record interview? I'd bring three short examples from the test

4:05 question pack. Let's have a look at this one here. Seven real jobs for Jupiter T6 Astra. 30 minutes on what the idea guy can delegate. Hi Sam, you told Mike Allen this era is the revenge of the

4:14 idea guy. My channel is that idea guy. Every week I take my newest model. I hand it on my camera and show prompts. Last week I gave GBT6 Astra and Claude Fable 5 Fable one. Blah blah blah. Lots

4:24 of information. I would like 30 minutes with you for one question. Do you know? I'll be honest. I think Fable 5.1 guys wins this one. Genuinely, I think Fable 5.1 is better with the email. I will say

4:36 that it loses it on the subject option, though. I don't like its subject. Um, this one is a little bit shorter. Honestly though, guys, I would say the they're fairly comparable here. And from

4:46 a token point of view, by the way, um, they about the same. 2.5 million tokens for Claude, 2.7 million tokens for Chat GP2. And if you're wondering on the website front, by the way, I'll be I'll

4:57 be fair here. It was 5.5 million for Claude. And again, chatb did did that in 4.3 million. So almost 20% savings. Now the point here is these aren't supposed to be like the best emails in the world.

5:09 Like if I had given it my email system, like how you create good emails, they would have done a lot. I'm purposefully giving them no context to see how good are they independently at doing this.

5:18 And independently without anything else, these are way better. And then we got questions that we can actually go ahead and ask to Sam. I would say honestly both of them are fairly equal. If I had

5:27 to lean one way the other, I would edge towards Fable 5.1 on this particular task. Now, for level three, we're going to be looking at image generation. Now, to do this image generation, all I

5:37 literally did, guys, is I plugged in Higsfield to literally generate any kind of images and video that you want to has all the best models out there. You can literally come over here, click on

5:44 images, generate them, you know, and literally from this upload it, do loads of things. And basically, I just connected these to make it fair via the CLI and MCP. And all I did is I

5:54 literally come down I screenshotted this I and I just copied it into chat GBT include and I was like hey connect them if you haven't done that that's how you set it up and then from that I can go

6:02 ahead and use them. Now the test that I gave them on this essentially was to create a thumbnail for this video, right? I wanted to go and have a look at outliers and build something that was

6:13 kind of interesting and the results I got back were quite interesting. So this is what we've got. Okay, now GPT Astra on the left and Fable 5.1 on the right. Now guys, it does not take a genius

6:27 person to figure out which of these two thumbnails is better initially. I think that we could all probably agree that the one on the left is a lot

6:37 better. It is. There's still some things maybe you'd want to do to it, but I think that yeah, that is by a long shot. So, Charg has really understood the brief of being able to find like

6:47 thumbnails that work and actually outlier thumbnails and doing it. And it did it in 2.7 million tokens compared to almost 3.2 million tokens with Fable 5.1. Again, still an outrageously

6:58 capable model. So, I would say GBT6 Astral wins the first concept. Second concept, who would you hire? I'm not in love with either of these two, but obviously this one is better. Um, this I

7:11 don't really know what's going on. That doesn't make any sense to me. I don't know why it keeps adding these things in the bottom. And then the third one, you know, that is clearly a lot better. I

7:22 think I know which video that's from as well. some things I do differently, but just its ability to use Higsfield, generate these images, and find something that you could use. GPT6 Astra

7:31 is quite far ahead to be fair. I'm I'm massively impressed with that. Fable 5.1 falling behind on this task, especially with like the the D. And I'm also going to give you my own lived experience

7:42 after these tests. I just wanted to give you some objective um systems so you know exactly what you're doing and what's going on with that. So, now we've done that. Let's go over to section

7:52 four. Now, section 4 is a really interesting one. Now, essentially what I've asked for here is a couple of short graphics explaining GPT6 Astra and explaining Claude Fable 5.1 just to see

8:03 how good at doing it. So, we're going to play both of these at the same time. So, on the left hand side, you can see this is GBT and on the right hand side you have Claude. Now, a lot going on with

8:13 both of them at the same time. I'd say both of these guys I would say are fairly even. Now, I know for a fact it can design way better graphics than this. For example, all these graphics

8:22 right here were built with GPT6. We got the Sentient, Signapse, Orbit, Resonance, even this AI constellation, which added those logos in. And as you can see, you can even ask questions down

8:32 here. I just asked it to make this component look like DeepSeek again. And GT6 Astra did that for me on one shot. It is pretty incredible. Now, it takes us nicely onto the next level, which is

8:43 a really interesting one. And what I wanted to do for this one is give it something that I actually use. So this is the Aentic operating system for Claude for Chat GB2. This is my memory

8:52 system. It has all my files, all my decisions, so I can consult it and get lots of incredible information. It has things in there like how much money I'm spending across all my different apps

9:01 and how to save. And really interestingly as well, this section here was built by GPT6 Astra in one shot, right? I can see my bank accounts, my income, room to breathe. I can come down

9:12 here and ask it a question like Advent, what is this quarter's focus? Okay, come down. And guys, I'm not joking when I say literally this was all built with GBT6 Astra. Again, I can ask questions

9:24 about it. So, I can connect this to different softwares like Mercury and it knew how to do all of this stuff for me. It has to think about those questions. I can come down, I can see cash flow,

9:33 where I'm at on a month basis. This is sample data, by the way, for the purposes of recording. But you get the idea, right? Shows you how you're doing from a growth point of view,

9:40 systematizes it, even explains your connections. And again, look at this. This quarter's focus is signing six partnership deals, blah blah blah. You can come down and get an idea. And by

9:50 the way guys, if this is all sounding okay, I'm speaking Spanish. I'm going to put a link down below for my full cord code master class. I'm also going to be releasing my chat GBT masterass. You get

9:58 understandings about how to get it set up, how to build beautiful websites, power features, memory systems, Hermes agent apps, building anything as well as full access to this entire aentic

10:08 operating system. I'll put a link down below so you can go ahead and grab that. So Asha went ahead and built me a release room. Added it in, gave it a cool emoji. Effectively is a means by

10:17 which to kind of manage my workflow production on YouTube, which is cool. So I can come down. I've just refresh this so you can get a clean look at it. I can add videos. I can basically track

10:25 everything. So scripts, claims, recordings, visual assets, add things to my to-do list, paste assets, update things, and get a good overview of what's in progress, what's ready. you

10:34 know, a kind of like project management system for my YouTube, so to speak, which I think is a cool and interesting way to go ahead and play around with that. Then over here, let's have a look

10:42 at what we've got from Cord itself. Now, what this tried to do is essentially a bit of a cost framework. So, hey, here's how much money you're spending per video. I think that's a really

10:53 interesting data point. I don't know how that one necessarily helps me to grow or be better, but you can see how it's added in here. I probably wouldn't run with this one per se, but it has added

11:04 some interactivity in the session which is pretty interesting. Ultimately though on this one I would go with CHT6 Astra for its edge in terms of actual usefulness in terms of identifying

11:14 something for my Gent. It knows I create content and try to address that need. Exactly. Going to go for level seven which is going to be footage that I would use. And again for this we've gone

11:22 ahead and we're going to be using Higsfield. For this specifically though guys we're going to be doing something on video generation. So, I wanted to go ahead and use the like seed dance 2.5.

11:32 Obviously, you can just do this in Hakesfield. You can use a cinema feature. You can come down, just go to video and start making some cool stuff if you want to, but I wanted to go ahead

11:40 and have these guys just do it remotely, which you can do. Super duper convenient. Now, what does that look like in reality? Well, let's check out what they did. So, on the left, we've

11:49 got GBT Astra, and on the right, we have Fable. So, we can make a decision there. Let's check out Astra first. So, basically placing down some footages, separating them out. Same workload,

11:59 cinematic, open, insert. Let's have a look. I'd say that was pretty good. That was cinematic. That was decent. And let's see what we've got over here with

12:07 Claude. Looking at two sheets. Put them down. Having a think. One blue, one purple. Oh, I It's interesting how this was kind of going on. It didn't really make sense that

12:21 they just started the way that they are. Although, fair enough, that's not too bad. And let's check this one out one more time. Two different briefs, separates them out, got two laptops in

12:31 front of her, and then they begin the test. Cool. I think honestly, first of all, pretty damn cinematic, I will say. But ultimately, I think that would be fairly equal. There was nothing jumping

12:42 out to me either way for that. And that kind of really gets to the point that I want to make with some of these models about who I actually think the ultimate winner is and how this then crucially

12:50 tests with my experience. And I also want to show you some interesting things in terms of the data on this. Let's take a look at what the data says on this. Okay, when it comes to intelligence,

12:59 essentially equal depends on the benchmarks that you're actually looking at. From a coding point of view, GBT6 Astra seems to be edging it out a little bit, 2.1%. On automation bench, GPT6

13:11 Astra is crushing it. Reasoning, you see a strong advantage from Fable 5.1. Quite a few points ahead here, but crucially cost per task. GBT6 Astra seems to be a lot lower. And again, it's really

13:24 important to understand this is API cost per intelligence index task. So, it isn't necessarily how expensive is the API cuz on the API, they're actually both the same. It's more about what's

13:34 our bang for our buck with the actual tokens that we're using. So, if we're using both these models, what does that look like? And even this graphic here, for example, was made by Astra, which is

13:43 which is great. My lived experience with these models is they're both exceptional. I'm finding Astra just to be it just intrinsically just seems to get some components. I find I have to

13:54 ask it less questions. It just seems to crush it. And it's tough because I feel like a kid in a proverbial candy store because they're both fantastic models. I've been leaning more towards Astra a

14:05 little bit in terms of how I've been using it. Now, if I come down and just show you some of the actual specifics, I specifically wanted to grab the tokens. And honestly guys, it bears out 53.3

14:14 million processed tokens with 25 million tokens on GBT6 Astra. It bears up it's it's a little bit more token efficient which is always a good thing when you think about it from that point of view

14:28 and it's quite helpful that it's gone ahead and done that. Now what else can we practically understand here? So here's my personal recommendation. I am absolutely using both models right now.

14:38 Fable 5.1 and GPT6 Astra. I am finding that I am increasingly leaning more onto GPT6 Astra. I would say that it won the test that we ran today. I wouldn't say that it was like, you know, first place.

14:51 And who the hell is this Trump who just entered the arena? They are they're like brothers and sisters. They're like they're very very very close. Very very close. I'm giving the edge now to GBT6

15:01 Astra. That is currently the guy, so to speak. That is the guy. It's a step change ahead. I do believe Jensen Hong when he called this artificial general intelligence. I do literally think that

15:11 it is currently the model is Fable 5.1 Trump change definitely not. Am I using both models? 100 freaking%. And here's the thing. It isn't about which model is better. It's like which model for which

15:23 task. And the key thing that everyone misses here guys is that like I gave these deliberately really garbage prompts. Okay? And the reason I did that is I wanted them to work under bad

15:33 conditions to get an idea of actually what does it really look like? I could use a model that is a lot worse than either of these two models, but if I use the right systems and prompts, I will

15:42 get a better result. So, the model, remember, is only one portion of the entire system. You need to use the right strategies, the right prompts, like in design loop. That's how I was able to

15:54 get a website to look like this in one shot with Astra. So, you're only ever really going to get its full potential on building websites like this if you know the right design systems and

16:05 prompts to do that with. And that's the kind of quote unquote catch that people are missing out. The model is just one part. You need to be talking about the prompts, the systems, the workflows, the

16:15 strategies to get the most out of the model. But if you're going for raw intelligence, what would be your best bet for a one shot? Astra. But Fable 5.1 is not too far behind. And so, now we've

16:25 covered this comparison. We need to actually figure out how do we really reach the artificial general intelligence level out of these models. I'm going to show you one of the best

16:33 strategies I've done and we're going to cover that together inside this video right