GPT-6 Astra Is INSANE – Is THIS Actually AGI?

0:00 [laughter] >> Wow, look at these characters for for a 3 JS result. All right, GPT-6 Astra's here, so you know, no. >> [laughter]

0:11 >> GPT-6 Astra has been released. Although, it was not exactly an all-at-one release, which is why for the introduction in this video, we're probably going to heavily speed through

0:20 this blog announcement post because by now you're probably pretty familiar with what's contained in this page. So, I'm very very very excited to test this model. From what I've been seeing

0:29 preliminarily in a lot of the kind of like slop games that people used to benchmark these things for some reason. So, basically some of the things that we're seeing here it excels at seem to

0:39 center around its spatial reasoning capabilities, things like creating CAD things. I've also been hearing that this is very very good at computer use. I don't specifically know how I can test

0:50 that from the Linux system that I'm using here. However, that may be part of a test and in additional testing video of this model or the upcoming Fable versus this because I think that'll be

1:01 something a lot of folks are interested in. And for that, I intend to do some extremely difficult tasks with the two of them. So, not things like this, even though we love things like this, but

1:10 they also have like this spaceship thing and I love seeing games showcased prominently in announcement posts for models cuz it just kind of feels like validation of how like some of these

1:20 tests that I like to do are actually appreciated by the labs, which makes me then feel better about showcasing these models with these types of tests. With that, we also have some other tests.

1:30 Now, interestingly, Deep SWE 1.1 basically at this point just seems like it's not necessarily a decent measure of capability anymore. Um you know, cuz

1:41 it's just Muse Spark 1.3, I believe, is outperforming all of these on this specific benchmark and I don't necessarily know that that's the case in

1:52 reality. So, really I'm not even going to pay much attention to the benchmarks right here. We have our announcement post, we have cybersecurity mentions, and one thing to

2:00 mention also is the pricing. So, if we scroll all the way down, it is generally contained in the bottom of these announcement posts. And this pricing is directly in line with what Fable 5.1

2:10 costs at $10 per million in and 50 per million out. So, it is more expensive. However, if these benchmarks are to be taken at face value, it is also significantly more performant than its

2:20 predecessor, GPT-5 6 Soul. So, basically our next step now, I suppose we should look briefly just at some of the pertinent information bits about the model that we would see in the developer

2:30 docs, and then we'll move into testing. So, for some technical bits of information about this model, it has an April 30th, 2026 knowledge cutoff, so pretty darn recent, especially in the

2:39 scope of what we're used to having. A little over a million context window, and then 128,000 max output tokens, which is a rather healthy amount. Input is text and image in, so it is

2:49 multimodal, but it can't seem to take videos and things like that right now. Additionally to that, I do also want to mention that they've talked about a pro model being available as well to some

2:59 pro subscribers. For this video, we're not going to be testing that. We're just going to be using Astra as it is set up in the ChatGPT desktop app. So, just to make note of that as well. And one final

3:10 thing I would like to additionally mention is they do have some model guidance right here in the developer documentation, where they actually give you some suggested prompting best

3:19 practices and things like this, where apparently it will work autonomously, but it can ask for more questions where previous models would have made assumptions. They actually give you a

3:27 snippet to encourage more autonomous work and not do that. And then just a bunch of other things right here for best prompting practices. So, just to make note that this is available as an

3:37 educational resource for folks who want to know specifically the best suggested way to prompt this model. But, being that I have the finest handcrafted prompts for our testing suite, I think

3:47 we can basically ignore this and move into the first thing we're going to do, which is of course the browser OS test. Now, we're a couple of preliminary mentions right here. Let's take a look

3:56 at my usage. Now, keep in mind that I have run this task right now on extra high. This is of course the browser OS test V2.7, and I have also run the C++ self-contained skate game, which we'll

4:07 take a look at after this. I have not seen either of them, so I will be seeing them for the first time, but I wanted to start off as soon as I got access to it, I just wanted to run a few things. So,

4:16 my weekly usage limit right now is at 96%. So, that's pretty darn good. And then there are also these banked resets as well. So, I would imagine even if we drain this fully down, whether it be in

4:28 this video or the head-to-head with Fable, I will probably be using these sooner than later. All right, so with that, let's just No, I don't I hate the in-app browser to such a high degree.

4:39 There we go. Okay. All right, I like this already because it's giving us like these previews of these games that like I just like the way it's done that. Even though

4:50 it kind of spoils them. With that, let's take a look at this. So, okay, it opens with our launch pad. I'm not noticing any sound effects right now, but that's okay. I will say the

4:59 live wallpaper actually looks pretty nice. The clock in the date and time is in the top right in multiple locations actually. So, we have one in military time right here, and then one in

5:11 normie time, as well as the date. So, maybe a bit redundant. I'm kidding. >> [laughter] >> Let's check the big one. All right, there's no right click, which

5:22 is an obvious omission to try to fool us in thinking it is not Bench Marks to my prompt. We have What is this? Oh, okay. Make it yours settings.

5:32 Afterglow, deep space. I like deep space, but I also like that. I like all of these cuz none of them are orange. Is it still moving? Yeah, but very slowly, and

5:41 then Oh, that's nice. It's like a breathing exercise meditation thing. Generate a new variation of this wallpaper.

5:50 A fresh world generated just for you. I normally don't look at these first, but that's okay. Can we minimize this to this Mac OS dock and then reopen it? Yeah, we can. Nice. All right. So far,

5:59 so good. We have a search function. Oh, and we have This has a lot of interactive elements aside from a right click. Let's check launchpad. All right. Let's just Let's just start

6:12 with the GTA game. I will full screen. Yep. That's darn good. I believe there are no mesh colliders on the vehicles, but that's okay.

6:22 I think this yellow Oh, yeah. Look. All right. There is no sound. But, that's okay. Package collected. Follow the delivery

6:39 marker. Oh, okay. I would like to explore the map a bit as well. This is not bad at all. It's It's pretty nice, actually. Probably one of the best ones. However,

6:51 I will say that I would have liked to have seen some pedestrians walking around to be able to interact with with the vehicle. And additionally to that, just maybe some sound effects. But

7:00 overall, it does seem to have some gameplay logic. It's smooth. It's well put together. And I think we're about to deliver this package and get some money. Okay. Not bad. No Oh, there are mesh

7:10 colliders on the cars. Cool. Are they on the building? Yep. And we even get a little red flash on the screen when we collide with something. Oh, I see a police car.

7:21 Oh, okay. So, that Good. All right. Not bad. I have to add the sound since there aren't any. All right.

7:34 Acceptable. And this was run on extra high. So, extra high is one below ultra, which we will test, but not for this. All right, our next game, of course, it is

7:45 some like space flight fight thing. Hey, not bad though. Cool. Cool. Afterburner engaged does work. It doesn't seem like we shoot, it just seems like we avoid, which is fine.

7:58 It's a pacifist variation of this game, which I'm cool with. All right, let's see what happens if we lose. We have our whole integrity up in the

8:09 top right, and that's going down. Seems like we lose a quarter each time. Okay, not bad. All right, we have our mail app. That's very good looking. Well, it's in the theme of this OS.

8:20 Again, subjective would be the word to describe my judgments of aesthetic things like this, but not bad. And there's always meta references in the inbox to specific apps contained in the

8:29 browser OS. Just something every model does without having been specifically prompt to doing, so. The offline client saves messages and sent messages to explorer@ether.local also arrive in your

8:40 inbox. To send outside ether, use the button below to open your real email app. Okay, interesting way of integrating it, and cool to see that it does give you the option to like

8:48 integrate it with your real email client. What else do we have? Notes. A space to think. Things to try. Export. Okay, and it does save it as a text file. Good. Haven't seen that in a

8:59 while. Files. Calculator. 87 * 6 482 All right, now we have our special feature. There is one more app to look

9:18 at, but I want to look at this. Moment one. Save moment. All right. Restore. Very nice. Interesting. I had mistakenly had caps lock on, and it still

9:29 understood the command. Sometimes, a lot of times, that won't happen. It will have to be exactly as it's typed, so lowercase only would work. So, that's some nice little depth there. Who am I?

9:43 There isn't neofetch, but I just figured, you know, all right, simple, not bad. The final thing, I believe, is this field guide.

9:51 Oh, it's just a guide to the browser OS. And the special feature was the time capsule, which we did see. All right, overall, it was very aesthetically pleasing, and something that folks were

9:59 not necessarily in love with, at least with GPT-5.5 was its front-end capability. 5.6, I don't know what the consensus was, but it does seem like six maybe a leap in front end. And for

10:11 reference, that took 26 minutes and 16 seconds on extra high. So, actually not horrifically long when considering the reasoning effort this is on. And that's something to take note of as well, is

10:21 how long it takes to produce these results. All right, so, the next thing I had done with this was the self-contained C++ skateboard test. It cannot use raylib, and it has to have

10:30 the early 2000s New York City map. It worked for 22 minutes and 12 seconds, and I don't want to scroll down here because there's a spoiler of the result. And I haven't looked at it. I've been

10:39 going like this just to navigate the UI here because I want to see it for the first time together. So, with that, we have right here the reasoning effort. Again, this was done on extra high, and

10:48 there's just a folder this is in, so we can go and play it. So, let's take a peek at this. And again, okay, there's still a preview there. All right, this is not

10:57 bad. Let's uh Okay, that's actually This is very good. I think.

11:09 There are sounds. Now, the reason I'm saying that is the actual effect of the pumping, which is the sound effects, the pumping of the skateboard

11:20 is probably the more realistic one that I've seen before with the leg movement and things like that. The character's well like rigged for this specific game, I will say. Now, it's hard because

11:31 obviously the Fable 5.1 result is going to be comparable directly in my head right now and that may have had some nicer additional pieces of detail such as the bridge, the Staten Island ferry

11:43 and the basketball court. However, the pedestrians in this one are very well done. Can we No, so we just go through them. Okay, so that's something that I liked better

11:52 with Fable was when you crashed into the pedestrians, it was very funny and it was like mesh colliders, but I suppose this isn't directly a head-to-head video right now. So, let's just judge this how

12:02 it is and it's very competent. I want to see how do we grind on a rail? G is for grind. Okay. So, we may have to manually initiate that. Yep. I think.

12:14 Sounds like a drum simulator. I will say though, it kind of lags a bit. Ooh, that was sick. Oh. All right, that needs to stop happening. It's actually kind of hard to play this.

12:29 There's like some level of skill required. I will say putting H to toggle the help menu next to J and K is kind of a frustrating thing because they're right

12:38 next to each other on the keyboard causing things like that. All right, this is well done and I'll say this is probably the best character movement in terms of actually pushing the skateboard

12:48 with a leg. I've not seen a better implementation of that before prior to this. So, that is nice to see. And again, it didn't take super long for that to happen. 22 minutes on extra high

12:57 reasoning effort is not bad and look how That's only about 1,000 lines of code. So, it seems like that's a fairly efficient implementation considering all the detail and things in that few lines

13:08 of code. I almost have to wonder. All right, so here's our updated version. Off the bat, we're going to notice a couple of things.

13:18 One, there's like floating power-ups in this game. Two, there is a soundtrack. Three, the detail on the buildings has been overhauled to such a high degree that it's almost like shocking.

13:35 All right, I'm going to turn the music off for a second. The graphical improvement here is actually really pretty impressive. These are probably some of the cleanest

13:45 graphics I've seen before for a C++ generation. Now, I don't know what these things are cuz it doesn't seem like we can actually pick them up. There is a basketball

13:56 court now. >> [laughter] >> Was that there before? I really hope not. The board model is significantly improved, too. Look at that. It looks

14:04 like a little tech deck board or just like a normal skateboard. Can we Can we bump it? Okay, still we can't bump into the pedestrians, but And there's now Was that a moving vehicle or

14:14 was just one rendered in the distance? This is a big big big step up from the previous version just in graphical fidelity. This is really impressive.

14:29 Oh, I hate that like Oh. Good. Good. Fable did that, too. Look at the way they go and like Do you think there are sound effects for them?

14:45 You can like continuously chase them and then they go back and land. All right, this is a a very very big improvement. And there's now like a subway.

14:59 Wow, in all self-contained C++. That's really impressive. That All right, good. Good. This is what I like to see. Wow. So, this is what it made before and we see it just basically overhauled it

15:12 in a large large degree in terms of like rendering and fidelity. So, very interesting. This is the subway FPS game. This is a slightly improved prompt where when a wave is cleared, the

15:21 enemies have to be basically delivered to us on a train to start the next wave. That leads me into the next part, which is the sponsor of today's video, Superbase. Now, Superbase is actually

15:32 what runs my Ultimate Play website. It has prior to them sponsoring any of my videos. So, to give a little bit of background information and what is Superbase, I have here a special guest.

15:42 I'd like to welcome Bijan. Hey Bijan, thanks so much for having me. Huge fan of the channel and holding this makes me feel like I'm an influencer. Thank you so much, man. I That really means a lot.

15:51 Of course. So, Superbase is not only the sponsor of today's video, but they are the back-end infrastructure for ultimateplay.com. Whenever a game is uploaded, Superbase handles that, even

16:02 tracking something like upvotes, number of plays for these games. Superbase is what is actually behind that live deployment right now, where we can go and play a game that's being uploaded as

16:12 we speak for this demonstration. And I want to say, thanks so much for having me. I'm a huge fan of the channel and you're going to be at that 100k in no time. Well, I really appreciate that.

16:21 Thank you again for the kind words. Hopefully, knock on wood, we'll be at that 100k soon. And again, thank you so much to Superbase for sponsoring today's video. If you want to see more about how

16:30 Superbase is actually powering ultimateplay.com, you can click on the link in the description or go ahead and check out ultimateplay.com to see these live games actually deployed powered by

16:41 Superbase as the back-end. All right, so the Subway FPS is done and I've been just covering the artifact window portion of the screen with something just to not get spoiled. So, basically,

16:51 I'm going to remove this piece of paper and we'll be able to take a look at this together. So, um Let's Let's just take a peek at this. My

17:01 next test for the 1v1 with this in Fable is going to be quite something. So, um there's a spoiler behind us in the uh background. So, with that, let's take a

17:13 look at this Ashford Street. Why did All right, so Fable called it Ashworth Street. This called it Ashford Street. So, Okay, as long as it's not like the exact

17:23 same. Let's take a look at this in Chrome. All right, we have this now on our own system, not with the like weird artifact

17:32 cloud thing. Okay. I'm not hearing sound because for some reason it auto muted this. No worries. All right. Let's check our different weapons. Okay,

17:42 a little bit too much bloom for my liking, but All right. Okay, so this is the cool part of this is

17:56 they arrive Uh-oh, the game has frozen momentarily. All right. Not bad. This is sadly not really Oh, there's D fighting on the

18:07 floor. This is unfortunately not as good as Fable. It's a actually like not as good as

18:21 Opus. That is actually cool though. Look at the way that the effect scales when we run versus not running. There's no like I feel like some Okay,

18:31 that looks good. Folks always say try to get in the train. Um I can't in this one and in the Fable one I couldn't either. Uh there were mesh colliders invisible in the door. So,

18:40 okay, reload effect looks good. This is good, but it's not great and I want great. I don't want good with this model, especially because we use the ultra mode with this. Now,

18:54 I think what I'm going to do is probably end up giving this a follow-up. The enemy models are not bad though. All right, let's see what happens if we lose.

19:12 That's cool. End of the line. The depth screen on this one I like very much. That is actually That's quite all right. Best experienced with headphones. Eh,

19:20 not a big headphone person. Never really have been. All right, here's what we're going to do. First and foremost, we're going to swap this to fast mode because why not?

19:28 This is about half health half as good as I had anticipated. All right, so I gave it some positive reinforcement while also telling it this

19:43 is about half as good as I'd anticipated it being. Let's go for a 10 of 10. You got this, bro. And just to give it some positivity. So, I'm basically just going to cover the screen now again with

19:53 one of the AI generated magazines that was in our old Fable versus GPT video. I think it was that. Maybe it was a different one, but that way I won't get spoiled by this. All right, so spoiler

20:04 removed. Here is our updated game with the higher fidelity that I had asked from it. Let's go ahead and just start it. All right.

20:14 We have a bit of lag, but oh yeah, that's significantly better. I can already tell. It's more grungy. Let's just look at some of the fine-grain detail.

20:22 Good. There was a flashlight, so I just turned that off so we could see a little better. Oh, yeah. Look at that. That is well done.

20:32 Let's check these things out. See, this is This is better than Fable in terms of the models right here. This is like actually

20:43 quite quite smoked. No. >> [laughter] >> This is very very good. Look at the weapon model change as well.

20:53 Wow, look at these characters for for a 3JS result compared. Now, I know there's always those folks who are like, "This is slop." Like, "Blah blah blah." Sorry, but you know, that's how it

21:03 sounds. I'm impressed partially because of the leap in capability to get this compared to where we were even a year ago. So, that is what is partially so absolutely phenomenal about these

21:15 updates is the capabilities have evolved to a high high high degree. And that's why I get so excited. Also, I just kind of like this stuff. Now, let's

21:25 swap weapons. All right, there's a bit of lag right here. No worries. I do have the old one running which I should probably should have dealt with. Okay, good. I will say the fable one I

21:35 think still had a better train because the windows were transparent and you could see in them and it was really like quite something. Wow.

21:44 This is like I mean the models here are the just wow.

21:54 I Yep, okay. So, if you right click, you can actually aim. But, it seems like I can't shoot and aim simultaneously. So, all right, no worries.

22:07 Look at that. All right. Let's Let's have some I think Yeah, it said cuz I was looking at some

22:21 of its chain of thought. It said light fixtures that you can actually shoot out. That is sick. But again, it took two prompts to get it to get this level of detail. Okay, first impressions on

22:32 this based on everything we've seen so far, I think it needs to kind of get like it seems kind of lazy on first result. Where like it should just have put this

22:43 out from the get-go and same for the skate game. So, maybe I will have the prompt adapted to reflect that. The night edition. >> Uh and then

23:07 Now, that was very, very good. The characters here were very well done. I'm extremely extremely impressed with this. Fable had the better train, but this had

23:17 the better character. So, if we combine the two in some sort of fusion model, it would be very expensive, but also good, maybe. Those so far have not Every fusion model

23:27 I've tested has been like Oh, okay. >> [snorts] >> All right. All right, this was absolutely stellar.

23:34 All right, I'm giving it the watch website test where it needs to have the beautiful 3D model of the watch, a cinematic panning camera effect in the hero section, and an exploded view, but

23:43 I'm going to add a little something to this prompt just to hopefully mitigate some of what we've been seeing. So, I've just added in basically the experience I've had with this so far where every

23:51 result has been pretty good at first glance, and then when we say, "Okay, now make it better." it becomes absolutely jaw-dropping like the subway game with the actual character models. That was

24:00 absolutely phenomenal. So, I've told it that here, and we'll see what we get. Oh, I hate All right, well, it's it's spoiled. So, Okay. So far, this looks very good. Now,

24:12 I would like to see it natively in a browser. So, there we go. Okay, an unhurried summer. A very white take. Pretty much every result for this usually goes for some form of black and

24:22 gold. So far, on first glance, this looks good. Is it showing the actual time in my locale, which is something I now know to look for thanks to folks in the comments?

24:33 No, because that would be If it is, it's like a few minutes off cuz it's 10:10, but this is showing like 10:08. So, okay, I don't know maybe just a This

24:46 looks good. This is the first time I've actually seen the little pins that hold the links into the band before. I've not seen that. Now, the one issue I'm going to notice is this

24:57 face here is separated it's too high off of the face plate. So there is still some artifacting there and issues, but overall it looks decent. Let's scroll down a bit and see. We also do

25:08 have a date counter. The best moments don't watch the clock, neither should you. Good. We have our different types. Cool, and then it just changes it in the hero section. Take a moment, scroll to

25:20 reflect whichever one we have selected from down here. Good. This is what I'm interested in seeing is this exploded view. And we have little screws for the back

25:30 as well. The movement is unfortunately it's a bit hard to see because while I am trying to drag this to orbit it, it still doesn't fully give us a view of the movement, but that looks fairly

25:41 detailed, I think. Is it actually moving? I think the gears are moving. It's just very, very hard to tell. All right. Overall not bad. It's acceptable, but I

25:52 will say that this glass never quite goes back to where it is. And then additionally to that, um no, that was about the only thing. So other than that, it's it's quite good. And this is

26:02 the first time I've actually seen the pins holding in the links, which you can add or remove to fit different sized wrists. So interesting attention to detail. And just for a quick usage

26:12 check, it's saying we're at 91% left. It's That still seems fairly decent considering we've been running a couple different queries for each test, as well as using ultra. Now I do believe there

26:22 was mention of another reset coming tonight. So keep in mind I don't know this number could change if a reset does land. So we'll try to keep on it as best we can. So next up we're going to do one

26:33 that I know a lot of folks seem to like, which makes me happy just based on reading through the comments. This is the Chrono City test, where it needs to show us a 3D scene of a city block over

26:42 a number of different eras. At the end of this prompt it always has says go all out. And for this we have swapped back to ultra mode, and extra fast is on as well. So I'll be very interested to see

26:51 how it does with this. All right, let's take a look at our Chrono City result, which I want to see in an external browser. All right, 1985.

27:06 This is the first time a model has not made 1985, just some like dark neon synthwave looking thing. Are there sounds? There have to be. It's probably a toggle

27:18 or something. All right, we have pedestrians. Some of them That one's holding a shopping bag. Bicycle models, video world. The

27:29 vehicles look pretty good. Lots of taxis here. Even taxis with different colors that are still taxis. Okay, some of the buildings I'm seeing are maybe oriented

27:38 a bit awkwardly. However, as a scene, let's just zoom out and take a look. I like when they have buildings in the background, even if

27:46 they're not as detailed. We have tooltips. Neighborhood landmark, weathered limestone. Okay, window air conditioners. I had not noticed that

27:54 actually. Make some noise. So, like a stereo-related piece of advertising. The fire escapes on these buildings look good. Joe's Pizza keeps its lights on late. Grab a slice before

28:07 the next rep- record shop closes. Okay. All right, so here's what I'd like to do because normally we start with 1945. So, let's just Cool. Did the buildings get taller

28:21 for 1945? You know what? It doesn't matter. Okay, this is where I'm not as thrilled, truth be told, because the car models are still overall pretty much the same. It tried to do I

28:33 think some older school car things just like with the bigger grills and things. The outfits of the people did change and I noticed like the actual store names have changed. However, overall it seems

28:43 like the city as a whole is not It's like basically the same core artifacts just styled differently to reflect the time. Okay, so here we see the vehicles do

28:55 have fins now in the '60s. Good, I like seeing that. So, they did change a bit. All right, interesting. I'm going to It's actually hard to get a good vantage point for the transitions here.

29:09 Okay, hello tomorrow. Discover the details. Oh, that's just click on the tool tips. The rooftop billboard space-age optimism. We have a heading even, which is not

29:20 something I normally see. And we have buttons here to zoom in and out, which is nice as well. A brightness slider. Oh. Day and night. So, in some ways a

29:29 brightness slider. Look at the lighting actually reflecting on the building and how some of the windows are lit up still and some of the ones aren't. Oh, okay. Well,

29:37 it's just more noticeable at night. That's actually kind of nice and the car lights as well. Nice attention to detail as we see right here.

29:48 '85 we did see as we had spawned in on this one. This looks like a really good scene just actually in the night time more so than in the day time, but let's look at 2005.

29:59 Good, we've got a significantly larger building at least in one front. The vehicles are a little different. We'll see if they have like a singular LED light bars like

30:12 the models sometimes include when we go into 2025 and beyond. Let's take a look at the outfits of the people, which I also do want to see in the '60s. Okay, I saw someone holding a briefcase. The way

30:22 they're all walking in unison is almost kind of concerning. It looks like the original Macintosh like ad for some reason. I don't know if that makes sense. 1985, I wanted to look at the

30:32 clothing in '65. Interesting. A bit more grayscale than I would have anticipated. All right, let's look at 2025. All right, so the vehicle light bars

30:43 didn't change how I would like them to, but the movement is not bad. Are the traffic lights actually changing? All right, not unless they're on an actual like real-time scale there.

30:54 And then finally, let's look at 2055. Okay, buildings are taller. Solar panels everywhere, green, and for the first time also the model has not just assumed that 2055 would be full of

31:08 neon things and cyber aesthetic. Interesting, these cars almost look like all like autonomous Waymo cabs, at least this thing's take on them. Let's take a look at the people and what specifically

31:19 the outfits look like and things. Okay, they all have backpacks on. And let's see how it looks in the night time in 2055. That's very pretty.

31:32 That's a 2055 I'm down for. Yeah, look at the reflection of the light across some of the solar panels here. Really not bad. Now, I hate to say this cuz like I want to get this test

31:43 out cuz I'm more excited to do Fable versus this because the types of tests we're going to be doing are not things like this. Well, some maybe. I almost feel like it could do better.

31:52 I'm going to hold off though cuz that was good enough. So next up we're doing the street eat test. However, we're making it build this game using Blender for the assets and Godot for the actual

32:01 game. This is a bit more of an extended prompt than we normally give them, but the culmination of this is to just basically make street eat using Godot and Blender. Now, based off some demos

32:10 I've seen floating around online, this thing is pretty much state of the art at using Blender supposedly. So I am expecting very, very, very good looking assets here.

32:19 But honestly, the ones that the sub the subway models that it made for the second iteration of the FPS were unlike anything I've seen with 3.js before. All right, so when I close this file browser

32:30 window, we're going to see our first look at street eat. I have not seen it yet. So, okay. Bulk shipping, I have no idea. All right, I think it's running well,

32:40 something was running because I heard sound. That's probably this. Let's mute this Chrono City thing and then we'll figure out how exactly to run this. Open project.godot and press F5. Okay.

32:52 >> [laughter] >> Like That only took 15 minutes? Okay, can I full screen it without like the cuz the cursor just You know what? I

33:00 don't care. I will zoom in in editing. Wow, look at this. This is I mean, like It's made its own style as well. Yeetmart

33:17 I just I'm going to wait. All right. >> [laughter] >> chain reaction >> [gasps]

33:47 >> As I continued to yeet, something took hold [laughter] of me. It was like there was a different region in the driver's seat. Clicking, yeeting, clicking, yeeting. The more I yeeted, the more I

33:57 sought to click to yeet more. The game [laughter] was an absolute masterpiece, but in the process, I may have lost myself. This is very good. It's even got like tool tips of

34:08 things. Yeet street and whoop's ave. And the things respawn. Decent game play music as well. The sound is

34:24 a bit painful AT THIS POINT. >> I I'M SORRY. I'M LIKE IT'S just that's excellent. That is really really good. This thing with Blender and Godot seems like the possibilities are quite quite

34:59 vast for those with the creative intent. And that it genuinely it did that in 15 minutes using Blender and Godot and our usage right now is at 87%. So it's not really using We have to keep in mind

35:13 that's a weekly usage limit as well. So but I don't care about it as much cuz we have these banked resets at least for now. So not bad and we have been using it on fast mode and ultra. So that's

35:22 actually I would say that's okay. So the last thing we're going to do at least in this test video is something that was a historic test and I want to do this for a reason that this is something you can

35:33 look back on many generations of model ago and there will likely be some result on the channel assuming I tested it for this to compare to. So we're asking it for a 3D printer simulation using 3JS,

35:45 something that can be opened in a single browser and is contained in a single script. This is something I used to use very prominently to showcase in new models that were like more state of the

35:55 art at the time and I think it'll be kind of fun to see how this does on this task being something that's being marketed as AGI and hypothetically if not like the best model in the world

36:05 right now. So with that I also want to make note that there will inevitably be comments saying, well like I want to see more intricate tests. As soon as this video is uploaded, I'm going to start a

36:16 head-to-head with this Infable 5 1 and those tests will be significantly significantly more difficult than anything we've seen here. So just keep in mind that's coming up. All right, in

36:26 about 10 minutes we got stratum visual print lab. All right, so what's uh let's just look at this in our native

36:37 browser. So far it looks good. It's a the bed is going through the Z rails. You're right. It's a mechanical collision I should have caught. Yeah.

36:47 Yeah, you should have. All right, so it fixed the one issue which was basically the issue we saw there earlier. Now interestingly there's some Z fighting on

36:55 the build plate which is a little disappointing. There is no threading on these Z rail screws and then additionally to that we also No, that was about the only thing I was

37:05 going to like obnoxiously complain about. Overall, I mean this is a significant increase in fidelity than what we would have seen previously. Although that's a pretty extended back

37:13 end here. This is a janky printer. Let's just see how it looks. Heating up. And I want to make note that we want to see if the spool actually does rotate.

37:23 Yep, it does. Okay, the Bowden tube is just kind of like yanked in there which you know, I could have done without but All right, this is good. It's going

37:30 properly layer by layer. One of the big issues is it's kind of hard to like move. Okay, slow down. There we go.

37:42 Is the screen showing the layer? It's I would like to know but I frame hidden for a closer look. Whoa. Current layer tool path shown. All right.

37:51 Can we change our filament? No, we can't but it's got like a bunch of other things right here. Okay, overall this is I mean I just wanted to do something like light and

37:59 simple at the end. Just maybe something that would be comparable to a lot of other older models so we could see what we got from this one. Okay, the screen does not actually show

38:11 layer but I just was something I was wondering there. All right. Overall, that is probably going to conclude No, probably. That's going to

38:20 conclude our first look and test of GPT-6 Astra. Overall, this model seems like perfectly tuned to do 3D games and things like that, which I really like.

38:30 We're not going to do a results overview for this one because I want to just jump into the Fable comparison. Overall, the results were very, very strong. My biggest takeaway though is probably the

38:39 fact that it produced these results pretty quickly, especially compared to what I would have anticipated. Now, in that we also have to remind ourselves that the fast 2x speed was enabled. But

38:50 even still, okay, say this took 20 minutes, or maybe this isn't the best example of it, but let's say like our Street Eat game where it used Blender and Godot for the assets. Even if that

38:59 took twice as long with less usage being taken down weekly, it seems fairly token efficient and I like seeing that, especially the quality of the results that it's putting out are fantastic.

39:09 Now, we ran it either on ultra or extra high, and our usage throughout all this is at 85% left. Keeping in mind, we ran it on ultra a lot, and it was also on fast for a significant portion of these

39:20 tests. So, that seems actually not bad. I think that's fairly reasonable. There is of course that other pro model, and I believe that has significantly tighter usage limits, but because this is what's

39:29 available to most, I want to just focus on testing this now. And with that, that is going to conclude this first look and test. As I said, there will be a significantly, significantly more

39:38 difficult test coming up of this and Fable in a head-to-head, which I'm going to begin filming right now and hope to have done sometime within the next 12 hours. So, if you have any questions,

39:48 feel free to leave them in the comments, and thanks for watching. Subscribe.