Unfortunately. Whoa. It's like the bus scene in Harry Potter 3. So, as I start this browser OS, sorry, I'm excited. I've also been up all night. Kimmy
K2.7 Code has been released. This is the newest model in the Kimmy family. Last time K2.6 code came out and then following that K2.6 came out, I do
believe. So, we should hopefully soon be also seeing a K2.7. Now, Kimmy coding specific models are
very good. This is open source and the weights were published immediately which is very exciting to see. This is arguably I haven't yet tested it but
just based off of past experience and seeing a new iteration this is likely going to be one of the strongest if not the strongest currently available
openw weightight model. So this is very exciting. So before we get into it please do feel free to
subscribe as I would like that 100k plaque. And now let's take a look at Kimmy K2.7 code. Now there really has not been a ton of information here. I
don't really see any specific announcement release post or anything. However, the weights are now here on Hogging Face, which is very exciting. And I
think one of the main takeaways about this model is essentially what's listed right here, where this
has improved token efficiency over Kimmy K2.6, reducing thinking token usage by approximately 30%. So, if this is giving us performance either equal
to or greater than K2.6 six with 30% less thinking tokens. That is a significant efficiency improvement. Now, an improvement leap like that was seen
with GPT 5.5 comparatively to its preceding models, GPT54, and it was able to do the same tasks with
significantly fewer tokens, which is just like awesome for efficiency sake and cost savings and things of the sort. Now, in terms of some technical
specifications and things of the sort, this is a 1 trillion parameter mixture of experts model with 32 billion active. It has a 256K context length.
And something else that's very nice to see is that it does have vision capability. So, some of our
tests can just be giving it photos and asking it to replicate them. Now, something very interesting is they're showing benchmark results for their own
specific benchmark right here. And this model is not really performing. It's closer to GPT55 and Opus 48, but those are still significantly better
based on this. The interesting thing is the fact that this is actually showcasing that on their own
specific benchmark. But something that I myself am more inclined to focus on is the leap in capabilities between the preceding model and this current
model which do seem pretty significant at least in this one specific benchmark. There are not a lot of commonly seen benchmarks shown here. So perhaps
those will be coming at a later time. But just from what we see right here, this seems to be a big
leap between K2.6 and this. Now, if we go to X where you can follow me, my username is at Bjambo and I changed it so it's now my name. One of the
things we see here prominently is just a chart showcasing the reduction in token usage, which is exciting to see. So, the gray ones are Kimmy K2.6 and
the blue ones are K2.7 code. So, I am specifically using this through a Kimmy coding plan. To be honest
with you, I thought I had canceled it, but I I don't think I did cuz I saw like I still had a subscription. Regardless of that though, the price for
this model, as we see right here, is 95 cents per million input tokens and $4 per million output tokens. Something else that's cool to see is that is
the same price as the previous model. So, it's just nice to kind of keep tabs on pricing movement
over time. Okay, so basically what it's done here is it's trying to autonomously test this and it seems like it's encountering issues with either the
generated script or with the actual environment in which it's trying to test this. So, I'm going to probably just let this ride and then when we get a
concluded result, we'll see. And something I should note about this specific test is that I just
ran this straight from within build mode. I did not initiate this in plan mode. I will change that for some of the more intricate tests, but for now,
let's take a look at our Kimmy K2.7 code. Neon OS. Okay, welcome to Neon OS desktop one. Let's check the big thing. Is there a right click? All
right, there isn't a right click, but I guess I'm okay with that because it shows this is likely not benchmarks to my specific benchmark. Nonetheless,
we do have a clock in the correct time of my local and our start bar. Okay, user local account. Clean but simple. I'm just going to run through these
apps one by one. Something I like to check with Notepad is if we click save, does it actually save
it as a text file on the system? Nope. But just the local memory and it tells us there. I'll accept that. Check all the functionality in the buttons.
Good. Next up, we have paint. Can I resize? Yeah, I can. It's so weird models do this. Even Fable 5 did this where I think it was Fable 5. The paint
canvas does not scale when you actually resize the window. I guess that's kind of a nitpick, but at
the same time, we want max performance. So, settings. Oh, what is that? Is that a water block? Nope, it absolutely is not. I was mistaken. The
oldfashioned Pacific Northwest background. I have not seen this in quite a while. It's like seeing an old friend. Next up, we have browser. Oh, okay.
Wikipedia generally works with this. Someone I'm sure will inform me as to why that is because I don't know
off the top of my head. Let's try Space Defender 3D. Good. Part of the course for what we get. Oh, okay. I don't know how I failed because spaces to
shoot. Oh, okay. Am I supposed to just I think if these get past me at all, I lose, which is just ridiculous. Let's try our calculator. Okay. 64 * 9 5
66 + 10. It's late. Well, it's early. Never mind. There's a pretty decent. Okay. I think the game
exe icon failed to load, but there's a good bit of fake file types here. I like the scroll aesthetic there. Can we actually create a new folder? No.
Okay. Does the file search work? Okay. Well, it's pretty nonetheless. All right. We have terminal help. All right. And now the big one. Neon City
Crime. Okay. Wow. What a It's just a like just leaving nothing to the imagination with that name. Oh,
okay. This is the slowest car I've ever encountered in my life. Can I get out of it? Be faster to walk. Speed 2 miles an hour. Yeah. Okay. Well, I
couldn't collide with one of these pedestrians if I tried. Oh, yeah. That All right. Well, more or less the car model is there. The buildings are
there. We do have defined roads and moving pedestrians. So, I'll accept it. Finally, I would imagine there
would have been some special feature. Switch desktops with Ctrl Alt Arrow or click the desktop indicator. Okay. Yeah, that just like it switched it.
All right. Well, I can't get that special feature to work specifically right here. And clicking the desktop indicator doesn't do anything. But that's
okay. For a first test, it's always nice to get warmed up with this. Next up, from within plan mode,
we are going to have it create the beautiful static subway scene using JS. And then following the successful completion of this, we'll have it be
turned into a FPS. All right, I may be cheating a bit, but it's generated it for quite a while now. The thing that it's trying to do is just run it in
a headless browser, check there were no JS errors, which there weren't, but now it's continuously
trying to actually take a screenshot of it and then look at the screenshot because it is multimodal. seems to be encountering some issues in the
pipeline of doing that. Nothing related to like the actual code quality. So, I want to just take a peek at this to see if it does actually load. Okay,
it does. Good brightness. Okay, this is very clean. Look at the ground is like Oh, that's the train. All
right. Something I'm going to notice on first glance is the scene composition. I It seems nice. We have individual tiles on the wall as well as on
this wall. Now, it did say it was going to put graffiti in here, and that's what we can see on the train. I believe that is an attempt at graffiti. We
have floating particles. The lights are there as well. The floor and the do not cross strip. West
Fourth Street. Okay, that sign is well drawn. Downtown local. And this is a pretty big station. The materials seem nice. It's just a bit dark, which
may bode well for our follow-up prompt. Can we go in the train? No, we can't. That's actually not necessarily commonly seen. Usually, you can just
walk through everything here, and there are actually separations to the train cars. So, I'm going to
send it the followup that just says, "Using this scene, turn it into an FPS in the traditional style of testing we like to do." So, let's just see.
Okay, we still have some Oh, yeah. All right. The movement's actually fairly smooth. We do have a very defined weapon model. I can't see any. Oh,
OKAY.
These enemies are extremely tough.
The particle effects when we do actually manage to defeat one are good. Let's check the Okay, so that's still pretty rare to see is when the bullets
actually make holes in the environment. something that is always like uh something that butters me up to a high degree. Oh, okay.
That's nice. All right. Overall, the scene is still kind of dark, but we didn't really tell it to change anything from that. So, it did properly keep
it the same. Something I'm noticing though is it's put mesh colliders on a lot of the static elements in the scene, which it doesn't. I don't normally
see that like in the in the same way that we can't actually walk through the train. We hit these pillars as well and we stop we stop there, which is
nice to see. So,
okay. and some of the enemy logic is a bit questionable in terms of like can we actually even defeat them, but overall it was a solid result. Next up,
we're going to be testing some of the multimodal capabilities of this model. I have initiated it in open code from build mode and I started it from
within this photos folder. So in here is a bunch of different angles photographed of this little 3D
laptop that I had created that was 3D printed. So the task here is to create an accurate 3JS model of this. Additionally, it should have a terminal on
screen that the keyboard can be pressed to interact with. This is something that I had also tested in the Miniax M3 test video. So, we do have some
form of reference point for what we get. There are 14 images. Need to identify device, maybe a Cyber
Deck portable computer, custom terminal. Let's read a first few. We have an image of a retro portable computer. Beige case screen hinging keyboard
with white and black key caps. Label 1 ghahz turbo clock. Need advanced 3JS rendering. And it's going to view some more images. I'm noticing some
quote unquote like caveman speak in the chain of thought here. Things like since file large and like need
picture of device. This is something that has been talked about a bit online in terms of actually having the model speak in a far simpler manner in
its chain of thought and they call it like caveman talk cuz it will just be like since file large like me write file things like that more or less and
I'm noticing some of that here which is interesting. It's something definitely like further reading u
if you find that interesting. All right. So, it does seem like this is working, but something I've noticed that's a common theme, and this could
entirely be system specific to my setup is every time it's trying to do the QA checks afterwards, it just takes far too long and like hangs up or
something. So, with that, we're going to initially take a look at this. Okay. Now I think I think we can
definitely see the specific issue here which is that a lot of this is just massively inverted. It did put a functioning bit on the screen. Let's see
if we can actually press the keys. Okay, good. So now if I can kind of Okay. And the typing did properly get there. Good lord. Maybe we should have
let it just fight through the image um functionality. Okay. And this would be the back of the system.
It does have closed lid. You know what? That's actually that's cool. So, open lid, close lid. The movement's nice. Toggle RGB. Ah, that would be for
the keyboard now because of this Z fighting that's happening. Also, the way that the key cap is written on the side as well is quite curious. Not
necessarily something I've seen before. All right. Toggle RGB doesn't seem to really do anything. And
then we have reset camera, but it does have some texture to it as it would because it was 3D printed. In the interest of time, I'm probably going to
stop this one. I again, there's not really much reference for this specific test. The only thing we ran this with was with Miniax, which did a better
job than this, but it didn't have like the open and close lid animations. And this also didn't fully
100% finish because it kept freezing up when it was trying to autonomously find ways to QA check this just by looking at it. So, I don't quite know
what to say. Now, I've swapped this into plan mode and just put it in an empty directory. We're going to give it the self-contained C++ skateboarding
game test. And this is the one that has the California Boardwalk aesthetic denoted. So, it's given me
some follow-up questions here, but it didn't present them in the traditional like you press enter to select one. So, I answered them. The only thing I
specifically told it was you can't use ray lib. Everything else that it asked about, such as can it be a freekate game. What about bail logic? I said
it's entirely up to you. Just make it look good. So, this is now on its second attempt of rewriting
the file because as we can see up here, if I can get to it, could not find old string in the file. Must match exactly including whitespace blah blah.
The old string didn't match. Let me check the actual end of the file. And then it basically says, "Okay, I need to just rewrite this from scratch."
The big problem is it's already done this. This is the second time that it had tried rewriting it from
scratch. And now it's happening again. So this is onto its third time trying to do this, which is a bit frustrating. So we may see our first glimpse
of this result pop up on the screen here. Not bad. The frustrating thing is that the boardwalk is not visible. It's just blue water. I'm going to take
a quick peek at it before it does something else just because of it's taken a lot to get here. The
walking is interesting. Okay, we can do tricks. Uh, that was like I don't know why there's like blood splatters coming out from under the skateboard.
This is a pretty nicely populated area. And something that I don't actually commonly see here is if we go forward and it's hard to see, but the
skaters actually, it almost looks like they're pumping, which would be the behavior to get the board to
move in real life. Can we grind? Okay. Nope. That was a bail. The kickflip animation's actually not bad. And I think we did get a score there or
something. That was very odd. The problem with testing this right now is I just did the fable tests which were just they almost felt like they were
from a different galaxy, if you will. All right, we'll allow this to keep going as it's still writing, but
I did want to get a preliminary look at this because it did have a bunch of issues getting to this point. Okay, it's opened it again now. So, that was
not me. All right, so we did get a final result for our skate game. My assessment here is it's probably not going to look much different than the
preliminary test we looked at. Okay. And we don't really see much. I just the blood splatters when we
land are really quite disturbing, especially because they seem like they're coming from the skateboard and not from the person. So, the stores are
drawn acceptably. The posts are drawn acceptably for the lights. Actually, we also have some trees there as well. And there are a lot of NPCs that are
walking with some level of animation, like movement of their legs and their arms and things of the
sort. Though, I'm going to say the actual point to where we needed to get to this was pretty bad. Basically, it took three tries for it to properly be
able to edit the file without failing as that result that I had shown the error that was happening. And we still didn't get something 100% perfect
though for not allowing it to use RayLib. I this is an acceptable result. Now, because I was a little
disappointed at the process of creating this skate game right here, I have gone ahead and just straight from build mode, I've given it a different C++
prompt that I've basically run maybe three times in total ever. It is to using three C's++ generate a 3D racing game with the following features. Then
we have a firstperson view from inside the car, a steering wheel that turns when the player is
turning with WD, low poly graphicals, something that looks like a vintage rally game. So, I've just started this straight from build mode, meaning it
will probably decide to use Ray Lib, which is fine with me because I just want to see what we get for another C++ test as the Skate one was not 100%
smooth. All right, so I had to step out for a second, so I didn't get to specifically watch this one
finish, but it does say the racing game has completed and it compiles cleanly. So, let's just let's just take a peek at it. This happens more often
than not. Okay, it's tough because on first glance it's just like, oh, this is garbage. Like, it's bad. But you have to factor in this is more
difficult. Had we just instructed this to perform this in a single HTML script with 3JS, it would look
significantly better than this. This is not necessarily simple, but something I'm going to notice is it's actually moving up and down as it goes over
the terrain at least a bit. That's something that is in line with what one would want with a rally game. There are trees. The steering wheel does
spin. Now, the reason I'm more soft on this than one may expect is because this specific C++ test is one
that I had only tested with at then like state-of-the-art models, and these were previous generation state-of-the-art ones. I don't remember
specifically which, but like this is pretty in line with what I had seen when doing that. So, this is not 100% horrible as it may look. It shows, you
know, some capability. So I did just want to do another C++ test and also I ran that entirely from build mode
because of some of the issues we had with the skate one when starting in plan and it just glitched a lot. Now for my next test, I have given this just
through open router, a front-end design test that I had only run with Claude Fable 5, which was to create a high-end website, basically this for a
watch company, but it had to make the 3D assets itself, and then do a cinematic pan, making it look
like some form of thing rendered within Keyshot. What we see right here is the Fable 5 result. So, let's take a look and see what we got from Kimmy
Code 2.7. And real quick, the specific prompt for anyone interested is pretty short and just listed right here. So now we can peek at what we got.
Okay, that's not bad. That is not half bad. At first, I was totally like upset because it didn't seem
like anything was going to load. But we can see here we do have some semblance of a fabric looking strap. It is panning around. And something I'm
going to notice is the actual table that this is placed on has some interesting artistic lines to it and things like that. I do see a second hand
there. Some of the hour markers are perhaps a bit sideways, but it also does on the face of the watch there
say slapis. And we can somewhat control the orbiting movement here, but not to a super high degree. Although I am going to say I'm actually impressed
with this because yes, we saw the fable result first, which was obviously state-of-the-art. The gap between the two is not insane when you consider
that this model is 12.5 times cheaper per million output tokens than Fable. We even have like the uh
the date there or the hour or something. So, I'm actually more impressed with this than I was anticipating. Okay. 316L stainless steel sapphire
crystal god. No photography, no stock assets. Okay. And then we even have the spinning renders for two different watch styles here. We have the ember
gold which is $5,400 spinning as well with slapus written on the face. This I will say is a respectable
result. Some of the images and things are a bit like wonky like these should be placed in a different location here on the cards. But this overall in
terms of judging it from 3D assets, I'm impressed. I'm extremely happy with this result. I'm also thinking about starting the slapest watch company if
anyone's No, I'm kidding. Actually, side note, I bought these empty watch cases for like dirt
cheap. And the thing is you can buy custom movements for them and I have custom faces that are going to go in them that I can't show on the channel or
on video. But so this is partially why I like this because I just thought it was cool to be able to do this. So these were really like dirt cheap
comparatively to what you get and partially what's inspired me to do this test. Just a quick side note.
So I am going to run one more test that was also run with Claude Fable. This was in my second Claude Fable test video where we're starting out by just
giving it this AI generated photo of a demolition derby game, instructing it to create a pixel perfect replication of this game with soft body crash
physics and just a playable game that is supposed to look as close to this as possible. So, it'll be
interesting to see how well we get to pixel perfect. I have initiated this just from within plan mode. So, we'll give it a full opportunity to plan
and then execute. All right, it's asked me a few follow-up questions here just pertaining to like what it should focus on and then the controls. It
asked like should this be a playable game or should it just be a scene where you can move the camera,
which I suppose is a reasonable question given that I didn't really say make a fully playable game. All right, so this is where I've just it keeps
failing when trying to check in the browser, but when it's doing this, it's more or less finished the results. So I'm very interested to see at least
what we get right here in terms of replication. Okay, it did a spot-on job of the UI there. Now,
something I'm noticing here that is an issue is the gravity is too strong. So, the cars basically just snap to the ground and then get stuck or
there's too much friction or whatever you'd like to call it. Now, it did do a good job getting the aspect of the like the camera view of this is a bit
odd. So, it did a good job with that, but unfortunately, it's just not 100% kind of where I would have
liked it to have been. though the menu bar appears pretty spot-on and multimodal coding from a photo like this is difficult. So, I've given it a bit
of follow-up here. Just basically saying the cars are stuck to the ground, ditch the fog, and then just make everything look better. The cars don't
look very good. I can't actually show the specific prompt that I had sent it as a follow-up, but one
could imagine. All right, it seems like it's made some changes to the game. Significant improvement, I will say. Now, frustratingly, the cars are
still not moving, but it did do a better job and change the model of the cars. I did also get rid of the fog. Really, the big issue now is the
movement is still not working. Did our health go down? No. Okay, it started there. So, that's one of the
frustrating things. Now, I want to see if the soft body physics work. So, we'll continue to allow this to fix. Okay, so I'm going to call it here. It
did get the cars moving, but unfortunately it's just not really This is not really making much progress considering the amount of time it's taken. So,
I I could have sworn I saw some deformation of the vehicle, but at this point, I think I'm going to
call this a fail. It's taken far too long, and I've given multiple chances. What is it going to I mean I'll let this continue running but for now I'm
going to move on to a different test. All right. Yeah, I went back to the to the low poly game and unfortunately it's just not like it's not 100%
getting it. So that's a little disappointing. I'm also giving this just through open router the Jerry's
apartment replication test. This is the one that includes the reference photo of the apartment from Seinfeld, telling it to create an accurate 3JS
model of this apartment. Still, no model has perfectly replicated the floor plan of this. They're usually just varying levels of detail. Okay, first
and foremost, the assets here are nicely done. That is I know this may sound crazy. This is actually one
of the better bookshelves I've seen before. Can we move around this with WD? We can. Now, this does not necessarily look like Jerry's apartment, but
the thing is really none of these results have yet. So, we'll judge it just based off of the individual assets and the composition. I would say this
really is not bad. We have the green couch. We have things on the fridge. Our bookshelf and the green
bicycle there again is one of the more bicycle looking shapes we've seen. We also do have the little Superman over there. Let's move on over. Some of
the wall placement is a bit odd. We do have a good-looking bed with a lamp over there. I think we have some additional Okay, that would be a bathroom
in there. This actually did a nice job on some of the individual composition of this scene. Not
necessarily true to how the apartment looks in the game in the game. I need to get off these 3JS games, but not in the show. This is overall though
not bad. Now, I have also given this a 3D printer simulation test just through open router. However, aside from the standard prompt at the end I wrote
I also need to be able to upload an STL and have it print that as well. Now, this is something that
historically I've saved for very high-end state-of-the-art models, but I figure because the 3D printer test is probably long in the tooth at this
point, it may be worth upping the complexity a bit. All right. Unfortunately, we have some blatant issues here in terms of the rendering of this
printer. I'm quite perplexed as to what I'm actually looking at. Okay, let's Oh, wow. That was only Let's
just try to upload an STL and see if that works. So, this is a small little remote control car engine block. Fantastic. That does actually work. Now,
we have issues with the way that the base plate is being drawn here and things like that, but the problem is this is almost such a realistic 3D
printer sim that it's incredibly slow because it's doing 336 layers. Though, I can definitively state that
is actually printing the proper STL here. So, that shows basically the actual like graphics here. The implementation of that pretty poor, something's
not right. the implementation of the actual slicing and printing of this STL is actually correct and spoton. So that's arguably perhaps the more
difficult thing to have correctly done. So I suppose we can give that a pass there. If I jump the layer
height up and then restart it again, it's it's difficult to see, but I actually the core point here is it did properly allow us to print an STL that
we uploaded, which is good. I don't know why, but I just want to try this. It randomly came to mind. I've sent it this image, which if you watch the
channel you're familiar with, and thank you, and said, "Create an animated SVG of this image." We'll
just see what we get. It may be entertaining. All right, let's take a look at the animated SVG of that image. Okay, it's not that bad. Do I see any
animation? Okay, there's some particles moving very, very softly in the background. Overall, it properly captured the scope of the scene. I would say
so. This may look funny, but it's actually it's really not that bad of a job. It did capture their
facial expressions quite well, too. Next up, I'm going to run some things through Open Router just in parallel. One of them is going to be, of course,
the flight combat simulator game. Then, I'm basically just going to duplicate this tab and run another one. Probably maybe the ship combat simulator
game or the motorcycle game. All right, here is our flight combat simulator result. Okay, very
interesting models here. Sky Combat. All right, it's playable in terms of like actually being able to fly that works. The models are a bit simpler.
I'll be honest, I would have expected better for the specific task, but keep in mind that we did also run it through open router, so it was not
benefiting from a proper agentic harness like open code or something of the sort. But even just off the bat,
I probably would have expected a little better to be completely transparent with this result. All right. I had given this a motorcycle simulation test
now. All right. We're going to see. Unfortunately, Whoa. It's like the bus scene in Harry Potter 3 where like Okay, this is this is not it. Uh the
speaker is on. Yeah, this is this is not it. I'm sorry. The the track actually looks okay. Like the
way the trees are drawn and things like that. Let's just try All right.
All right. Uh, yeah, I'm gonna Not quite. So, the final test was going to be the drum kit sim, and I gave it quite a bit of time, and it just
unfortunately is not working. So, that's going to lead us into the conclusion here. Now, I have to measure my expectations comparatively to having
just tested Fable, which genuinely is like a different class of model. anyone who's saying it isn't just hasn't
properly tested it. And I'm going to be blunt about that. So, I went back and looked at my Kimmy K2.6 model testing video, and I will say that in the
tasks that were cross referenceable with what we just did here, this actually is quite an improvement over Kimmy K 2.6, which is nice to see. However,
I can't help but come away with the lingering feeling that it feels slightly undercooked. Now, I
did ensure 100% that every sampling parameter is correct. it is properly integrated just based off of the Kimmy documentation as well as the sample
references and hugging face and everything like that. So, it's wired up correctly. I just noticed that some things just didn't go together as I would
have hoped they would have. Kind of like the drum kit simulation and things like that. And this was
perplexing the multimodal coding we did where it had to create the laptop. The lid open and closing animation was actually pretty cool and it did put
a functional albeit kind of flipped over terminal here and we can click on these and there's some Z fighting and things but that is overall pretty
darn cool. Additionally, the thing that I was arguably most impressed with was the website for the
beautiful front-end slappis watch collection where it actually had to create these assets and I would say it did quite a decent job here. Something in
line with that was also impressive was this 3D printer. Now, the printer itself was quite a disaster. However, we did also tell it that it needs to be
able to print in STL that gets uploaded, and it absolutely did function there. It was just a bit
slow, so we didn't really get to any higher layers, but that is a 3D model that was made in a previous test video, and it did actually start printing
it properly. So, the logic for that was good and that's nice to see as that's something I've historically reserved for more like pro models and things
of the sort. Additionally, what else did we have? Okay, this is not really applicable. That was
just kind of like an entertaining thing. I found that the flight combat simulator and the sport bike racing game were not very good. They should have
been better and I don't quite know what to make of that. And then we also did have a couple C++ tests such as the skateboarding game first and
foremost. And this is one of the ones that I went back and looked at my Kimmy K2.6 testing video. And this
was a significant improvement over the previous result. Just the board animations, the NPCs, the actual layout of the map and everything like that. It
was much better than the previous version, which is very good to see. And it did also do because we had a bit of issues actually getting to this point
where the result was concluded. It tried twice to basically like rewrite the entire thing after
hitting a failure. And then finally on the third time it actually successfully wrote everything out in chunks and it was able to run. We did also have
the rally racing game, which on first glance may not look great, but it actually did do a nice job in having the steering wheel move around as well as
the driving and like moving around up and down the terrain and things of the sort. So, overall, I
don't really know what to say as a takeaway. It's a definite improvement over its predecessor just based off of me having cross referenced some of the
examples with the previous test video. Though, I will say I'm going to be interested to see what folks experience with using this. And yeah, I guess
that's what I wanted to say. And I have to keep in caution that I did just test such an insane model
that the bar was shifted a bit. So, testing this right after may not have been 100% fair. And I want to be honest about that. So, that's partially why
I'll be very interested to see what folks come back um in terms of this. Also, of course, we had the Subway FPS game, which I totally forgot to
mention. Here we go in subway station once more. I mean, this was pretty cool. This in particular was an
extremely big improvement over the previous Kimmy K2.6 test. So, that was cool to see as well. And really, that's probably going to conclude today's
video, our first look and test of the newly released Kimmy K2.7 code model. If you have any questions, please feel free to leave them in the comments.
And thanks for watching.