GPT-6.1 Sol Is HERE – Can THIS Beat Claude Opus 5.5?

0:00 Check this out. That is absolutely like XG Games mode pro skate. Another day, another Frontier model. GPT6.1 Soul is here. Today we're going to be testing the newest model of the day, which is

0:12 OpenAI's GPT6.1 Soul. This comes quite soon after GPT6 Soul, which truthfully did not seem to have the best reception in our testing. However, that doesn't mean it's bad. It just means it failed

0:25 to really wow us in terms of some of the 3D centered results that we like to run on the channel. So for today's video, we're going to be testing the newly releasleased GPT61 Soul. OpenAI had

0:36 their dev day today and they did introduce a ton of additional things. However, this video is going to be solely focused on Soul. So before we get into it, do feel free to subscribe and

0:46 let's start out by taking a look at what is new with this model. So, the biggest takeaway from this release is that they mention this is near Astra intelligence. Astra being the currently best available

0:56 OpenAI model, which is $50 per million output tokens and $10 per million in. This is near Astra intelligence at a fifth of the price. So, really, this model is competing in price with Claude

1:08 Sonnet 5.5, which as we saw yesterday is a surprisingly surprisingly capable model. So, I will be very interested to see how this stacks up. We can see right here on the benchmarks that show how it

1:20 compares to its predecessor which is GPT6 soul. It does seem to be a quite decent improvement in capability for just 8.1 increase. Now, interestingly, we see with the deep SWE score, it does

1:32 the best here on high reasoning effort. It is becoming more commonly seen that the models aren't necessarily producing better. Better meaning more efficient results on higher reasoning efforts such

1:44 as max or ultra. However, we're probably going to do a lot of our tests on max because I am interested to see just the maximum capability of the model. And sometimes the degradation in score for

1:55 certain benchmarks at a higher intelligence comes from the model doing more than what is assessed or judged in the benchmark. So, keep that in mind as well. Now, for a bit of technical

2:05 information about the model, we can see right here pricing is the big takeaway. $2 per million in and $10 per million out. This is again the same exact price as Sonnet 5.5 which was a very very

2:17 capable model. I don't believe that with GPT56 and Sonnet 5 that Soul was a direct Sonnet competitor but does seem like now at least based on pricing that has come

2:28 to pass which is interesting. We do have an April 30th 2026 knowledge cutoff a little over a million token context window and 128k maximum output tokens. input modalities are text and image only

2:41 which is fine for a model of this class. And with that really I think our next step is to just hop into the GPT app right here on the system where we have our browser OS that is still working. I

2:51 am running this on max effort. It's been working for about 19 or 18 minutes and let's take a look at our usage prior to really concluding all these tests. So for the beginning of this test, our

3:02 usage is at 98%. Now, keep in mind that I am currently running the browser OS test as well as the pool party GDAU test which was introduced as a new test yesterday in the sonnet video just to

3:14 make sure that the impressive results were not the result of just benchmaxing commonly seen tests. So, with that, all right, so this is interesting. I ran this on ultra and I did run it on fast

3:24 mode. Maybe I shouldn't have done that. This was the backyard pool party test which was first introduced yesterday in the sonnet video. Now, I can tell you on first glance, this is not as visually

3:36 impressive as what we had seen with Sonnet. I made sure that all the cache folders, all the scratch folders, everything like that were completely removed. So, there was no prior

3:44 information about any of these results. Let's just take a look at this being that it is running in GDAU right now. Now, the positives I will say is this produced this extremely fast and it was

3:54 on ultra mode, so it could have spawned less capable sub agents to do this. And I do believe it's probably prudent to redo this, but I did just want to see what it did.

4:04 Let's see. Okay, hold to charge. Not bad. The water effect isn't bad. Interesting. We have like this off-white and orange aesthetic here, which is becoming the new like AI slop tell in an

4:22 online UI. Hold to charge. Okay, again, it's this is a respectable result. I think the biggest problem is this may not be a fair comparison being that I ran this on ultra and also on

4:37 fast mode. I'm actually going to have this redone just on max without fast mode on and we'll see how it differs. All right, in 33 minutes and 42 seconds, we have Halo OS. So,

4:50 open with Google Chrome. Good. All right. So far, so far I dig this. It's simple. Okay. First thing off the bat, if I were to nitpick, some of the application icons

5:04 are not appearing. That's okay. I do like this initial section right here. Just this like command center, if you will. Okay, we have our files there. Let's check the big thing. There is no

5:17 right click, which is an obvious attempt to not appear benchmaxed. Oh, we can full screen. I'm pretty sure that breaks recordings when I do that. So, one second. All right, so we do have the

5:27 date in the top right. And I believe this is the wallpaper setting. Okay, not bad. I do like that it is very Mac OS 10.5 style, which I'm a big fan of. We probably will I mean like leave it on

5:39 that. And then we have adjustable seeds, which is something that has been included pretty often recently, and I do like that just to add a bit of variability. All right, let's click on

5:48 this. Okay, cool. That's actually not bad. It does have like a little dock or command center. We have Launchpad. Oh, okay. And that was what was open initially.

5:57 Was that a teaser of our GTA game? Let's just take a look at it. Okay. First impressions, it's very the color palette is muted, but that's

6:12 not a bad thing. Can we uh Okay, it doesn't seem like we can interact with people. E is to enter. Well, a we do have

6:32 Oh, okay. And then that that small model smell when the police just snap exactly to your location and it's not put in any logic there to be like, hey, there needs to be like 10 units of distance from the

6:44 location of the police cars to the player car. However, this is not bad. It's a decently put together city. I will say nothing here is bad. It's just again it's going to be

6:56 very difficult to not compare these results to the sonnet ones being that one this is the same exact price and two that was literally like 24 hours ago. So it's tough to judge these independently

7:07 but I don't even know if I should be. This is a Sonic competitor. So you lost the patrols. The city is yours. All right. Can we take this one? Oh well. And then

7:17 this baby's got some torque. Next up, let's check orbital run. Good. I actually like this. I This is nicely done. And yesterday's test, we didn't even get a second game, which was

7:29 on me because I reached out and stopped it after 2 hours. So, time-wise, this one definitely was a lot quicker. And this was also on maximum effort. So, I do like how the sounds are actually

7:39 going up in pitch. Beautifully flown. Thank you. This was actually a rather enjoyable game. What happens if we Oh, okay. Not bad. I actually liked that mail. And these are just meta references

7:52 towards like apps on this. Can we actually launch the game from the app? Yeah, we can. Okay, that's kind of cool. Just from the mail app. So, they always try to wire the mail app into relevant

8:02 things in the game. Flight complete and it just gives us status reports in our orbital run. That's not something the prompt denotes. This is like some behavior that all models have converged

8:11 on doing where the mail app is related to some of the games and things like that, which is kind of cool just to see when they converge on different things across different models from different

8:21 manufacturers. Can we save it as a TXT? Okay, we can. I always do like that. That always bothers me up a bit. Sound Lab. Hold on a second.

8:40 No. Oh, yeah. If it had put like a bass thing in there, I would have gone in. All right. I liked that. That was a very nice

8:54 inclusion. I always do enjoy when there are like sounds or things like that included. Connected atmosphere. Your wallpaper responds to speed and police heat in games. I did not notice that.

9:09 Okay. I don't know that I saw the wallpaper react, but that doesn't mean that it didn't. It just means I didn't see it. This must be our special feature. Keep a moment. Revisit a world.

9:19 So, it's one of those like Chrono features, which is always a good thing. Like these are cool always. So basically, save a capsule, save moment, and then if we close that, and we

9:29 restore this, it will bring us back to where we were. All right. Overall, definitely competent, but nothing that blows me away. And were we to compare this to Sonnet 55, it was not as good,

9:42 although it was significantly significantly faster and also run on its highest reasoning effort. So next up, I'm going to try that pool party game again. I've removed every single

9:51 reference to it, all the cash, everything. It's not just archived on the system. It's gone. And this time we're just going to run this on maximum effort without fast being on. Prior to

10:01 this, we did it on ultra ultra code or whatever this reasoning effort is. I forget. I get them confused between this and anthropic and without fast. This is just for a simple overview. Using

10:12 Blender and GDO, create the following game, Backyard Pool Party. It is essentially a 3D backyard pool party where the goal of the game is to dive into the water and various like

10:22 characters and things of the sort. And this is using Blender and GDAU. Just something I wanted to add in because it's totally new. So, the models will not have seen this before. I swear I

10:31 just went and checked the specific like everything it did, everything it searched, everything it found to ensure that there were no remnants of the original results for this that it made

10:42 when we first started testing that we saw that was done on ultra with fast mode enabled. It basically made this like the same exact thing just slower. So again, the the Sonnet result was far

10:56 more visually impressive. I would say, especially the characters and stuff. I mean, these are pretty simple characters. They're not bad, don't get

11:06 me wrong, but again, it's really difficult to test this when having these Sonnet 55 results fresh in the mind. I mean, if we look at this right here, this is not necessarily bad, but like I

11:16 would like to direct our attention to what was created here with Sonnet 55. And like I mean, that's uh there's no comparison there. I'm not even going to sugarcoat it. That is like a an absolute

11:28 moppage of the floor um compared to what's seen here. So, we're going to let it finish 51 minutes. We have our pool party result. And I swear I like there's no prior art on the

11:42 system from the first time it did this. I got rid of everything. It's just really quite similar. Unless I messed something up, but I'm pretty sure I didn't. I was looking through all the

11:51 tool calls and seeing like what specifically it looked at, what it looked for. I mean, it just it's just really similar to what

12:01 we got. I mean, which I guess makes sense given that it is the same model. All right, let's just big splash energy. Okay, let's just go with Biggie.

12:18 Hold space to bounce. All right. Again, it's really like this is a solid result. I believe that we can Q is to flip, E is to twist, and space is to tuck. All right.

12:34 Okay. All right. All right. All right. One more splash. Oh, it's not bad. It really isn't. I think probably the biggest issue here is just

12:58 that it's coming after the Sonnet 55 result, which was which was better except for when the people would just get heated and land on the pavement. That was quite disturbing.

13:11 So, I mean, definitely a simpler take on this, but I wanted to do something that was not at all possibly benchmarks. Next up, we're going to be doing the self-contained C++

13:22 skate game. This is the one where it needs to be an early 2000s New York City block. It cannot use ray and I have taken a suggestion from the comments again because it was an excellent

13:32 suggestion where basically it needs to have a replay feature. So the traditional prompt here is the same but at the end I have appended there also needs to be a feature where tricks have

13:42 a replay option which gives a cinematic replay of a trick triggered by pressing Z after performing a trick. So, basically, if we do a cool trick, we press Z and it should give us a

13:53 cinematic replay of that. Again, something I saw suggested in one of the recent videos, and it's a good way to keep the tests a little fresh, but also keep them something consistent so we can

14:03 compare results across models. First impressions, aside from that's another test. I'm going to have to just wait till that one finishes. But

14:17 first impressions, this does not look bad. I can't help myself. I need to at least see like Okay. I mean, I think one of the perhaps main issues here is the um

14:30 arm is not I don't know. I will say though, the graphical fidelity here looks like what was produced with a previous version of this model after telling it that I want it to go harder

14:42 and make this look better. So, this almost looks like our second iteration of this test with the previous GPT6 Soul model. So, so far so good. The block is actually pretty clean.

14:54 I don't see any bail logic. Okay, we'll just just restart there. All right, cool. I'm trying to just do like a sick trick so then we can check the replay. All

15:10 right, let's just try the replay. Cool. So, when we press Z, that's our replay feature. That was a fantastic suggestion. So, thank you to whoever had left that. Very well done. Something I'm

15:20 going to say that I'm a little disappointed with is just this is a pretty bare bones like block. Now, it does say it just needs to be a New York City block. However, there's no

15:29 interaction with pedestrians or anything like that. Sonnet definitely mopped for the floor with it in terms of like a lively city. However, that is cool.

15:45 You know, I just I want to see what comes from sharing this. To be honest with you, I'm actually genuinely struggling with filming this video because if the model's just not that

15:56 good at 3D stuff comparatively to the clawed ones, that doesn't mean it's worse. It means it's worse at the types of testing that we predominantly like to do on this channel. So, I'm having a

16:06 tough time because if this isn't its strength, I don't want to just make a video dumping on it and being like, "Wow, Sonet is so much better. It's possible. This thing might be really

16:14 good at like optimizing like edge devices to run something efficiently." But, I'm genuinely really like trying to think of something different to showcase this with that's outside of like the 3D

16:25 stuff and more in line with maybe like lower level software capability. So, I'm almost a little frustrated. It said it's going to make it better while keeping the single file C++ requirement. It

16:37 doesn't necessarily say it needs to be contained in a single file. And I'm like 93% sure that the Sonnet 55 result had a bunch of different files. So I want to also tell it that because that's

16:49 possibly limiting some of its capabilities here. So I see this go all out. All right. So, it it redid the result based off of what we had showed it, which Sonic 55 did. It's

17:03 definitely a significant significant improvement. I'll say the water is quite well done now. Good. We do have sounds. The graphical fidelity is a bit improved. Although, it doesn't seem like

17:16 the city's very big. No, it's not. Okay. So, it just made it look really good from this angle, but it didn't, you know. That's okay, though. We do have some ships and things

17:28 like that. Can we grind on these rails here? No. I think we might be able to, but you have to like really time it good. That looks pretty good. That is an

17:47 aesthetically pleasing replay. and it does get the whole trick. So that's a whole another aspect of this is it has to determine when the trick begins, when the trick ends.

18:00 Okay, that check this out. That is absolutely like X Games mode prokate. That was that was awesome. So this still is not at the intricacy of the result that we saw with the Sonnet 55.

18:18 Did I call that fail earlier? I don't remember. But it's definitely an improvement just in terms of the water. The view that it started us from was basically this right here looks

18:28 incredibly impressive. The water is really, really nicely done. The reflections of the buildings on the water and things like that. That's cool. We still can't interact with pedestrians

18:38 and things of the sort. I didn't tell it that, but it was again something that we noticed in yesterday's test. So interesting to see, but the replay cam makes this really quite awesome.

18:55 Okay, I was going to see if we could replay it. Um, it didn't work going into the water. So, significantly better. So, I did the robot arm test. Of course, this time I

19:05 put it on high effort, which is the one that was at the highest performance in that benchmark chart on Deepsw S. I also turned on fast mode because for this test it makes sense. It worked for 18

19:16 minutes and I will say the car got moved a lot of times. It got flipped on its side, moved far away. One thing that I was impressed by was the fact that each time the car got moved, the arm moved

19:27 really well to its position. So, it was doing a very good job of navigating towards the car with the arm. And unfortunately after around 18 1/2 minutes I had stepped out for a second

19:39 and I went back into the room to see the arm just like in a full vertical position just stopped there and I went and read its message and it basically gave up. I'm not going to knock it for

19:50 that because while the claw result did successfully complete it with Sonnet 55, it did it horribly. It was like moving itself in destructive ways. It was basically luck that it did do it. This

20:02 one was far more nuanced and had better movements, less destructive movements, although it was lifting itself off the table a few times, just a slight bit, but overall it gave up. That's not

20:13 necessarily It is a failure in this test, of course. But the fact that it knew like, you know what, I can't do this. I'm going to stop. I do give some respect to that decision rather than

20:24 just like I'm going to crawl across the table and grab this car, which is kind of what happened yesterday and was just weird. But again, it just was so funny to walk back into the room and see the

20:33 arm just like like this, like what happened? So, an interesting result. Next up, we're going to do the subway FPS test. And this is the one with the slightly changed prompt where instead of

20:44 the enemies being delivered by train, we actually have to get on a train when we clear a wave, which brings us to a next station. Again, this was from a suggestion as well, and this made it a

20:54 lot more fun, at least to judge and play because then we get to go in the train and ride it for a little bit of time. So, we'll see what we get here. So, in just under 46 minutes, we've received

21:02 our Subway result. I had just told it that I need it running locally, not in one of these chat GPT sites, just because that way it's easier to open in Chrome for me where I'm not logged into

21:12 my chat GPT account. So, let's take a look at this. From our first glance there, it did seem to be a fairly respectable result graphically. So, okay. Yeah, that is. See, that's I

21:28 like this a lot so far. I'm happy with what I'm seeing. Let's just explore. Look at the reflections from the wet floor, which is just like permanently wet. I guess that's okay. Very well

21:40 done. Right click site. All right, we do have some sound. Me turn it up. Let's check our shotgun. Oh, and remember this is Let's take a look at these enemies. Not bad at all.

21:53 This is the one where we have to board the train to go to the next station. So, all right. And now the train is coming. Often times in results from like any of these models

22:07 when the train shows up, this is very nice so far. Look at the uh the bendy section of the train. This is good. I like what I'm seeing so far. This is definitely more in line

22:18 with what I had hoped to see for this. The train does have a bit of movement. However, as with Sonnet, these things are not swinging, so pretty bad for the realism. No, I'm kidding. All right,

22:28 let's see. Are there bullet holes in the environment? I forgot to check. Yeah, there are. Good. Good. The front of the train looks nice as well.

22:38 And we can get in like any of these cars. Walk to an open door. Press E to board. Oh, okay. We We had to Only the middle

22:50 car is something we can get on. Nice. I should probably reload, huh? All right. E is to leave the train. Oh, okay. And

23:04 it auto did that for us. Seven hostiles keep moving. The train just backs out of the station, which I'm not huge on. That's kind of akin to the police cars like being directly on top of you in the

23:15 GTA clone. This is just clean. I mean, all right, we'll go to one more and see. And all the stations are basically the same, which is okay, I guess.

23:39 And we have a larger enemy now in the form of this one. All right, let's see what happens if we lose. Okay, you made it this far. Not bad. That was actually quite well done. I

23:52 think I have no real complaints about this. It was competent. It was clean and it did properly follow instructions. Also, it was pretty quick in 46 minutes. So, I'm satisfied with this result. Next

24:04 up, I want to do a test that we have not run in quite a while. This is the replication of the apartment from Seinfeld. I've slightly changed this prompt so it does not denote it must be

24:13 done in 3JS being that this system does have Blender on it. This is a prompt so far that I think the best result I ever got was actually from GPT55 in codeex and that was because we had to turn it

24:27 into like a beat him up and it was just absolutely fantastic. It was fun to play. However, no model has really accurately tagged the layout of this as it is in

24:37 real life. So, often times they'll make good-looking scenes, but they're nowhere close to what the layout is as shown in this photo and as at least assessed by watching the show. So, we'll see how

24:46 this does. I do think just based off of some initial looks at this model and testing, I think its vision capabilities may be quite strong. Now, additionally to that, I do just want to show that

24:56 result that I was speaking of. So this was the result that GPT55 had made and yeah it's super simple but like back when that came out and it made this the fact that you could win all the other

25:09 elements are like shaking and there's fun effects and things like that. This may seem pretty simple now looking back in retrospect which is wild because this was not that long ago and when this came

25:20 out this was like wow this is absolutely phenomenal. So, think of that leaping capability. But I just wanted to showcase this as one of the results that stood out to me as time has progressed

25:29 as being one of the most fun and exciting. And that's on my website, Ultimate Play. So, if you want to go there, you can see it. I haven't really done as much with

25:38 this as I wanted to, but um hopefully in due time I will. And that was April 2026. So, I mean that's six months ago. four to that's five months ago and six days. All right. So, in about 40 or so

25:56 minutes, we received Jerry's apartment. This, I will say, is absolutely phenomenal. The individual elements here are exceptionally well done, and this represents a huge leap over what we

26:06 would have previously seen. Just looking at things like the actual multiple locks on the door, the cleanliness of everything, the individual models, this is an exceptionally well done result. It

26:18 is still not true to how it looks to us watching the show or how it looks to us in the reference photo I opted to give it. With that said, judging this based on individual details, it is very, very,

26:31 very well done. We even have things on the screen of the computer. What is the deal with airplane peanuts? New material. Thursday, open mic at 8:00 p.m. It went in and put like really

26:42 insanely impressive detail and thought into this Easter eggs, if you will. We can see out the window there are like other buildings and things of the sort. If we look at these elements

26:53 individually, this is a very very impressive result just from building a piece of scenery I suppose could be said. We have the green Klein bike. Very well done. Even detail into the chain

27:05 and things like that. We can go in the bedroom. There is a mini map there. The Porsche poster. So it almost seems like it pulled real assets to put these posters and things like that to make

27:14 them true to how it is in the show. This is just fantastic. And we have presets for different locations. So the office, the bedroom, dollhouse, which is just this in like

27:29 that dollhouse form. This is a very, very good result. No model still has done it absolutely true to how it looks to us. However, this is really really nicely done.

27:43 We have set details. Very cool. And it's actually sided sources here as well for the set details. We also have a lighting control as well. Warm table lamps. That's rendering quality balanced. So we

27:57 can even change that. And where we see that gets much simpler there. We can put it on high window daylight. So that should adjust what we see over there by the window maybe.

28:10 Oh, okay. No, that's just adjusting the lighting based on how much perceived daylight would be coming in from the window. We also have the TV a baseball um frame on the wall. So on ultra, I had

28:22 run the same Runescape 2007 prompt or maybe a similar one that we had done with Sonnet. One sentence saying, "I want a pixel perfect replication of Runescape 2007 PvP at the Grand

28:33 Exchange." It worked for 19 1/2 minutes. The first result looked pretty identical to what we're about to see. However, the chat window was a bit buggy and then the it was too zoomed out. So, we now have

28:44 that here and it is running locally in our computer. So, check this out. This is like basically the UI elements right here and everything. The actual icons for these weapons, the sharks, the

28:56 potions, basically everything outside of like right in here is truly truly pixel perfect. Well, okay. At least in the in the inventory screen right there. This is exceptionally exceptionally well

29:09 done. And we can zoom in just a bit. Now, that is probably my biggest complaint here is it's still a bit too zoomed out and the players don't look as realistic as they should. However, it's

29:18 even put in some of the NPCs we're used to seeing right here and the way we can walk around. Can we move with the mini map? Okay, we can't. Now, do speech bubbles appear?

29:29 Sadly, they don't. All right. I seem to be having a bit of difficulty trying to attack people. Ah. Oh, okay. We do want sound. Good. And it's made the name of some

29:45 folks who are like uh well known in these games. All right. A All right. and then we lost. So, let's

30:06 go see if we can find someone that is a lower level. All right, it doesn't seem like we're going to be able to, but still, this is always a fun one to run just to see how they do. It's definitely

30:14 got its audience because if you've never played this game, this probably doesn't make as much sense. However, overall, I'd say this was a pretty well done result. I can't seem to enable special

30:26 attack there. Can we move the camera with the Okay, good. The arrow keys do move the camera. Is this how Runescape looks to like people who don't play it? Like this is

30:37 probably how we look if we're sitting there like pking and getting really into it and then someone just externally is looking at it and sees this. All right. All right. We got to do some

30:47 damage here. Come on. Oh, what the We're not able to special attack. So, that's a big issue here. All right. This is ridiculous. Let's eat to full health and then let's

31:00 just take care of this guy. All right, it's time to sit. Oh my goodness. Yes, GF. Oh, it it auto like says something kind, which is definitely not

31:17 in line with what happens in reality. And then we got some simulated loot. Okay, overall that was definitely Oh, V is for vengeance. All right, hold on real

31:28 quick. I'm going to just try something. All right, good. Interesting. Overall, that was well done. Ah, okay. And we could have full screened it

31:38 there. Um, maybe I'll just zoom in in editing. This is pretty well done, though, from a UI replication here. These are like so pixel perfect that I'm wondering if it just kind of clipped

31:49 them from somewhere and then used them because that's how good it was. Yeah, these are these are dead on identical to what we see in the game. So, I think it may have done that. definitely um not

32:00 that but pretty cool to see. Now the final thing I did which I did do on fast mode in max effort which took around 45 minutes and I just sent a follow-up saying I'd like

32:11 this to be running locally on my system not in a chat GPT site. I asked it for a high-end 3D Blender model of a Saub 900 Turbo and then also a web page where a user could go here and interact with

32:23 specific parts of the vehicle like opening the hood and things of the sort. So, I'm gonna say this does not look identical to one. Part of the reason I choose this car, aside from just really

32:34 liking these, is because it's such an odd unique shape for a car. They're just weird cars. So, it's harder to replicate them. It can't just use like a traditional sports car shape or

32:44 something. So, this is definitely not quite in line with what it is. However, some elements of this are more or less acceptable, like that little piece right down there. And then the structure of

32:55 this front lower veilance is actually better than you may expect. Additionally, the headlight textures and things of the sort are not bad. The grill, we have an interior and we do

33:05 have a hatchback and some other things. This bumper back here is also pretty real for like what these have as well. These little strips right here in the bumper are a telltale indicator of this

33:16 vehicle. So, that's a nice thing to see. Let's uh Okay, we have auto rotate. Let's open the hood. Very good. Now, the fenders don't come up like that. The hood's separate to them, but the engine

33:27 model and things like that are actually quite detailed all things considered. We even have shock towers visible. We have a battery. Now, these are not necessarily in the right location. Like

33:38 in real life, the battery is over there where that is and things of the sort. However, it's fairly detailed and the rims do look very similar to what these actually have. So, so far this is it's

33:50 nice even if it's not 100%. Lower front windows. That's cool. Stuff like that. I like to see headlights. Nice. And then we can change the color as well. Very good. I I like the red

34:03 best. So, we'll do that. All right. Let's take a look. Good. It is a manual. Now, this interior is not really how it looks in reality. But again, like with the engine, it's just nicely put

34:14 together and nicely done. I mean, this angle right here is quite respectable, I would say. And then the interior looks nice as well. The shift knob, the steering wheel, and this angle, like

34:23 right here, is actually somewhat impressive. These vents here are things that the car has as well. So, overall, while this is definitely not like a pixel perfect replication, this is a

34:33 respectable take on it. And we can also take the driver's seat. Very cool. This could be used for a game. And then we have return to the exterior. And we have presets for specific views as well. So,

34:46 just something else I wanted to see just basically to test some of its 3D modeling and design capabilities and things like that. Now, the final thing I did, which unfortunately isn't going to

34:55 properly make its way into the video, is that I gave it an old Pentium 3 laptop, some old Gateway that I had purchased from some flea market, probably the MIT flea market. I don't even remember how

35:06 many years ago. I think like probably north of 10 now. I've never used it really. It has like Red Hat on it. So, I plugged it into this with a USB connection and I was like, "Okay, I want

35:18 you to do something with this." Now, unfortunately, it was not able to establish communication just from USB to USB, which is not the fault of this model. So, then I found the laptop

35:28 actually had some little networking card in that would allow it to use an Ethernet cable. I then did that. I said, "Wait, I found," which is kind of like how these things talk in the internal

35:36 monologues. I found this LAN card. I plugged an Ethernet into it. Now the link is connected. Can you check again? And then it asked for a couple of things. I should also mention there was

35:45 a camera set up. So this could actually physically see the screen of the laptop and what was on there. So then I had changed it to the Daybreak model, which is one that is a little more pmissive in

35:55 terms of the boundaries for like cyber tasks and things like that. I didn't realize that that would downgrade it to GPT6ole and not use 61. Additionally to that, it unfortunately was not able to

36:06 get into the laptop that was just booted to a login screen for Red Hat in any of the open ports. It actually did check and the culmination of this was I would have needed to do some manual work

36:16 whereas my hope was if I just had an established link, this could basically work its magic and get something done to reinstall any new OS on the system. So unfortunately I stopped that and it was

36:28 just basically the whole reason I even thought to do this is because the model was not as impressive as Sonnet 55 in the visual tasks that we were doing and I was really struggling to find

36:39 something totally different totally not a visual task to see how it performed on that and this didn't really work out very well but I wanted to showcase it because one it is going to contribute to

36:50 whatever our usage is at for the conclusion of this test and two to show like I really am trying to find other ways to test these that are not just like a sick Blender model. Though I

37:01 would make the argument that I do think some of those capabilities of making 3D models, games, and things like that can translate into a reflection of the model's overall strengths and

37:11 weaknesses. So overall with that, let's take a look at our usage throughout the entirety of this testing. So the entirety of our testing today, our usage is now at 95% left. I believe it was at

37:22 98% when we started. So all of this used 3% of our usage. Now I need to make it clear that I don't know if this $200 a month plan has been h haved yet and that is something they said that this plan is

37:34 going to see half of the usage that it would have seen previously. I don't know if that has been put into effect yet. However, this really did not use our usage very much and that's nice to see.

37:46 So, I will say it does seem to be pretty token efficient in terms of what this figure ended at based on where it started. And we did do a large amount of tests, some of them on ultra, a lot of

37:55 them on max. And I would say that's actually respectable in terms of how much usage this used. So, overall, that is probably going to conclude our first look and test of GPT 6.1 Soul. Overall,

38:09 I don't want to make a very definitive stance on this model based off of a lot of visual 3D tasks, but I'll say I'm not as impressed as I was hoping to be, especially following the test of Sonnet

38:20 55. Normally, Soul, well, normally, I mean, for how long these things have existed, which really isn't that long anyway, Soul would have been thought of as an Opus competitor. However, if

38:30 Sonnet is the same exact price as this model now, and based off of what we saw today and what we saw yesterday in the Sonnet video, this is kind of putting out results that may not be as

38:40 impressive as Sonnet visually. It's basically tough to judge this following that test where Sonnet was just absolutely mind-bending. And this perhaps not so much, but it may have

38:50 strengths in other unexplored areas. So, I'm going to continue testing this in things that are not fully game based and more softwarebased and like physical devices. So, that's my closing thought

39:02 on this. Overall, it's competent, doesn't seem to use a lot of usage, and it is relatively cheap. So, that's going to conclude our first look and test. If you have any questions, please feel free

39:12 to leave them in the comments. Thanks for watching.