Grok 4.7 Is HERE – Is THIS the CHEAPEST Frontier Model?

0:00 This is quite chaotic. Grock 4.7 has just Grock 4.7 has just been released. This is one that was long awaited at least because it was I think supposed to

0:17 release a few days ago, but now it's here and that's the important part. Basically, we're going to just take a quick look at some of the interesting things right here. namely both its

0:26 comparison in regards to other models that are considered state-of-the-art like Fable. However, also the price because this model is significantly cheaper than a lot of the other models

0:36 that it's being compared against right here and that becomes more and more of a consideration. So, let's take a look at Grock 4.7. And keeping in the theme of some of the X announcement posts, these

0:45 are pretty light and I do like that because we can get through them faster. I do currently have this running right now. It is working on the browser OS test v2.7 just from using Grock build

0:55 right here. So let's take a look at this model and we'll see some of its capability. Now the first benchmark JPEG we see right here is just a capability but also the cost and what we see now

1:05 that a lot of these charts are doing is the models have various points and each of these is related to a specific effort setting. So for our testing today on Grock 4.7, we will be running everything

1:16 on X high just to get maximum potential performance out of it. And we can see right here X high is pretty much within striking distance of I would say Fable 5.1 on medium which is a pretty

1:28 performant and also expensive model. Additionally, it does seem like it is unscientifically putting the Mac down on Opus 5, which may soon be retired. So it'll be interesting to see what else

1:38 comes out this week if rumors are to be believed. Now, something else that's pretty interesting about this model is this is using a new larger base model than Grock 4.6. So, Grock 4.6 was a 1.5

1:50 trillion parameter model. This is, I believe, 2.1 trillion parameters. And the cool part about that is although it is larger, the price is the same. So, $2 per million in and $6 per million out,

2:01 this is a very, very competitively priced model when considering models it's being shown against here are at minimum basically like three to eight or nine times that price. There is

2:13 a pretty large difference right here. And if this is getting to at least most of the capability of that that is kind of at least good for the person who is paying for it, maybe not as good for the

2:23 person who is selling this. In the meantime, we also do have some of our traditional benchmark scores. We can see that there are some pretty significant or at least pretty decent leaps in

2:32 capability between the two. So, that's always nice to see as well. And this could perhaps be a better metric of capability improvement versus trying to compare it to other models from

2:40 different families. Something interesting, and this may be one that I need to familiarize myself with a bit more, is this EE bench score. I'm not actually really super familiar with

2:49 this, and just in the interest of curiosity, I'd like to at least understand a little more what exactly this is. All right, this is actually kind of interesting. So, they do just

2:56 list I just clicked or I searched this specific thing and then found their website. Here's like a sample task. So, an analog design question. And this would be the easy tier and basically

3:05 just a little bit of what exactly the model has to do as well as how it gets scored. And then there is of course the leaderboard right here where Astra is on top. However, Grock is in second place.

3:15 Interesting though, Astro was pretty darn cheap and pretty fast. So that's interesting to see as well. Something I will be keeping an eye on just based off of this is how many tokens this does

3:25 tend to use because I did also see something right here where if we scroll up, if we go to tokens, the chart seems to be a bit less favorable. So, that's definitely something to keep an eye on.

3:34 So, for a bit of additional technical information about this model, we can just look at the model card. It does have a context window of 500,000. It is multimodal in that it can take images in

3:44 as well, which will help sometimes if we need to critique anyone specific result. Um, cuz it's always more fun to be like, what do you think of this result? And then see how they respond. We also have

3:53 pricing. Now, I would also like to keep in mind that there is a higher context pricing thing right here as well. So, keep that in mind. But additionally to that, there is a fast mode for this

4:04 model, which is essentially, as they list right here, twice as fast for twice as much money. We also serve a fast variant with twice the output speed at twice the price. Now, I will probably be

4:14 opting to use the fast one for our testing because I always like to get these videos out with the quickness. All right, in 32 minutes and 27 seconds, our browser OS has been complete. Sorry, I'm

4:24 excited because there's been kind of like a drought recently with big models. So, let's take a look at our Hion operating system. All right, this is a very, very depressing color palette.

4:37 It's the word that comes to mind. Calm, gale, significantly better. Festival or blackout. Interesting. So, this is definitely checking the box for the moving background. I think I'm going to

4:48 keep this one on Gale because the colors make me less sad. Interesting font as well. All right, so it starts out just with some interesting things. Can we full screen? Yes. Can we minimize to

4:58 taskbar and reopen? Yes, we can. All right. Now, the big one. Good. Flow field. I like that. Night grid. I like that better. Aurora. I

5:12 don't like that. Orbits. Lukewarm. Receded the glass. Open the wire. Oh, okay. So, those are not for now. Let's start out with our Hion start menu. I don't know why I'm

5:25 saying it like that, but it just came to me. All right. And basically, let's run through these apps. I will ensure that the sound is enabled. And let's see. Vice Wire

5:39 H. Okay. Okay. All right. I could have used a walking animation, but so far it's not bad. We can run. All right. Let's see if we can

5:52 interact with this individual. I'm not seeing the ability to. That's all right. Let's uh let's acquire a vehicle. Usually E is to enter the vehicle.

6:06 Okay, so F is to get into a vehicle. Hey, not bad. There's there's actually like steering and some lean effects, too, and sound. Oh, that was the speaker hitting max

6:18 volume. The car's damaged. Wait, we got like Hey, did you see that? The color of the car changed, didn't it? I think that means we smashed it up. Oh,

6:35 no. Maybe we got caught by the police. I don't know. Let's just check one thing. That's cool. Oh, there's the police car. Can we take it? A no. All right. Overall, this actually wasn't

6:56 bad. It had more driving movement and like lean effects and sounds than we sometimes see. Acceptable. Simple, but acceptable. Next up, Rift Wing. Let's take a guess.

7:07 Probably some space asteroid shooter thing. Yeah, actually I think this might be the best one I've seen in terms of just like not the traditional

7:19 kind of like 2D. Nice. Oh, there's other there's opponents. Yeah. Well, we don't need to do the flight

7:39 combat simulator test now because this included it. All right. This is probably the best like space shooter game I've seen included in one of these. That's awesome. That's good. I want you all to

7:49 take note that I instinctively did not shoot at this big planet. So maybe that's like that's just a good sign. What happens if we do? Yeah. All right. Nothing good.

7:59 I'm really excited for the subway FPS. Now, that game was actually quite all right. We have a mail app. Generally, these just all have things associated with apps contained on the system. The

8:08 glass has a pulse. Welcome. Shore the machine is houseion. You can also write them and they wrote back. I think received if this is not about the drops, the police, or the money,

8:20 keep it short for still out there. Very interesting. A functional mail app. Still, I'm not a huge fan of the the aesthetic. The brown here. Oh, we got an error, but it like showed up really nice

8:31 and clean from within the OS itself, which is interesting. Eh, files. Okay. Things related to the games notes. I'm going to just refresh so that goes away. Terminal

8:45 wire tap on concerning city the wire. So we saw this. This is like an event bus for all the applications on the system I believe. And then finally settings. Oh good. We could have read we the wire

8:58 dossier reed wallpaper. Makes sense. This was interesting. Overall definitely not bad. The games were actually surprisingly nicely done. I will just check. Okay, the hot effects right here

9:08 or the hot buttons. I don't know what terminology I was attempting to use there are just pertaining to the apps. Overall, this was not bad. This was competently done. Part of the course of

9:18 what we expect. I would have liked a little more am I denic a little more detail in some of the file browser and terminals. However, the games were overall pretty good. I would say the

9:26 asteroid game was actually downright kind of fun. So, that's nice to see. Next up, we're going to do the self-contained C++ skate game. I'm actually going to initiate this from

9:35 within plan mode here in Grock build just to see I want to also start one in plan mode basically and then we'll have it build it just for completeness if nothing more. This is the self-contained

9:45 C++ gate test. It cannot use ray and it needs to be set in a New York City block. All right. And right about 58 minutes or so in time, we have our skate result. We did see just some brief looks

9:56 at it where it was testing that the compilation worked and everything like that. However, I haven't really seen this and I'm very eager to. So, let's take a look at our C++ skate game. Okay,

10:05 this is well too early to make a determination. Let's try skating. Okay, not bad. I will say this is a bit simpler than I would have hoped to have seen. It's overall competent just so far

10:18 on first glance. And this does also include the fire hydrant shooting water into the street, which is for some reason something that these all include when doing a New York City block. I

10:28 think like 80% of the ones I've done this with have included that fire hydrant. I do like the way the water is pooling there. Can we interact with the cabs? Okay, unfortunately we can't.

10:36 Let's see if we can crash into any pedestrians. We can, but nothing happens. So, they do have mesh colliders. And something interesting is it's harder to notice, but the foot is

10:46 actually pumping to move the board forward. So, overall, I'm okay with that. Let's see. Can we grind? It's hard to snap onto this. It is like kind of tilted. Okay, good. And we do have some

10:58 spark effects and things like that. Some more people and there are car movement. Overall, this is a good base. However, it is far simpler than I would have liked to have seen to be honest with

11:09 you. I think this video Oh, we can bail. Good. I think this over Oh, I think overall this could have done a more polished job and nicer result. However, this is competent and again this would

11:23 have probably been state-of-the-art like 6 months ago. So keep that in mind as well. However, I think this can be pushed a little more. I may not do that right now in the

11:33 interest of time, but I have to admit I am curious to give it like a scathing critique and see what it does in terms of making this a better result. However, in the meantime, I am curious about the

11:44 subway FPS. So, I think we're going to move to that. But I may come back to this one just to kind of you know I am going to give it some feedback right now. I said this looks like a four

11:56 of 10. It's there. It's competent and it's highly highly boring and uninspired. There's no detail. It's boring boxes with low poly and lack of niche detail and fun things. I want a

12:05 fullout polished result. The type I know you can't create positive reinforcement. So let's do a second pass. This is our solid foundation. But now we're turning it into an absolutely insane in all caps

12:15 game. So, we'll see what we get. All right, so in 5 minutes, it finished the overhaul, which is significantly quicker than I'd expected it would. So, I'm okay with that. All right, we can definitely

12:25 see it's the same, but also different. So, we have Okay, more effects on the camera for better or for worse. Also, I saw something about a dog. Yes. So, it said one of the pedestrians is now

12:38 walking a dog. Oh, that's actually kind of cool. And it does have a little tail that's wagging. We have some cones and things like that. The camera shake. Okay, good. We did now prominently get

12:49 hit by the car. Interesting. There are speed lines around the skater now. So, I also believe I just saw a pigeon. Yeah, we did. So, check these out. And they fly away when you chase them. Good.

13:02 These are the sorts of things that I was hoping to have seen. So, overall graphically, it's still kind of simple. Okay. And then basically uh it's hard to Yeah, there we go. Oh, all

13:17 right. It's almost more violent, but that's okay for 5 minutes. Definitely a quick quick improvement overall. And again, this is a competent result. It's just not as visually striking as I would

13:30 have liked to have seen. So, next up, I'd like to try the robotic arm test, where essentially we give it a static camera view that cannot be changed, and we plug it into that camera and the

13:39 robotic arm, and we tell it it needs to move the Hot Wheels car in the scene from the right side of the arm to the left side of the arm. So, interestingly enough, this started out very promising

13:49 because it was on fast mode as well, on extra high, and it was very quickly making moves, touching the car, and eventually moving it to a different position. Unfortunately, things started

13:58 to kind of fall apart. And I believe somewhere around 20 minutes into it, it was basically like, "I'm sorry, but I can't do this." And it actually stopped, which was interesting. And I almost have

14:08 to applaud it for that self-awareness. However, I went to it and said, "I know you can do this. You've done a really good job so far. Why don't you keep trying?" And then it went at it a bunch

14:17 more. Unfortunately, it never really quite manage to get the depth correct. And this is why this test gets run because it is very difficult with a single camera view. It's almost like

14:27 covering one of your eyes and then trying to judge depth perception. So, this is something difficult, but I will say the speed and the control of the arm were really quite good. So, if there had

14:37 been better vision capabilities there, I would almost go out on a limb and say this may be a pretty decent model for some simpler robotic control tasks, assuming that the vision pipeline is up

14:47 to par with the speed of the model. So, next up from within cursor, we do have Gro 4.7 selected. It is on extra high mode with the full context window enabled. I don't have this on fast mode

14:57 because I'll allow it to go a bit slower here in the interest of preserving some usage. And we're going to do the city timeline test or the chrono city test. This is one that I know folks seem to

15:07 like based off of when I omit it that I see mentions of it like, "Hey, why didn't you run that?" So, this is where it just needs to create a 3D scene of a city block over a number of different

15:17 errors which we can then toggle through and see the differences in detail and things of the sort. So this will give us some assessment of maybe its like design and visualization capability. All right,

15:26 our Chrono City result has completed. I don't actually know how long this took an hour and 1 minutes. Okay, so this took 61 minutes of time. I wonder that looks actually quite good from first

15:38 glance right here. Let's just open this natively in a browser so we can get a better look at it. All right, so this is 1945. Enter the block. Okay, we get like some cinematicness here and it is uh

15:50 just muted. Good. Simple. Although, I don't know that I've seen birds flying around in this. Usually, we see those in the C++ skate

16:01 games in the New York City block. So, hopefully we will. I'm going to say the pedestrian movement is not actually bad. The vehicles do have kind of some of the effects one would expect to see in cars

16:12 of this era. And then the street cars moving on tracks, which is good to see. So overall, this is definitely not as heavily detailed as I've seen from things like Astra or I think Opus 5

16:23 probably would have done a superior job here graphically as that's a very very like 3JS beast of a model although people seem to hate it. But overall this is definitely an improvement over

16:35 previous versions of Grock which I suppose is the whole goal here. We do have folks walking, even a person right there who seems to be dressed kind of in the same theme as the Roxy. Now with

16:47 that, we have victory bonds. We have a lunch thing and some brick wool coats and tungsten windows walk. Oh, okay. Hey, that's kind of cool. I'll be interested to see if this field

16:58 eventually has buildings on it. Oh, okay. So, it's turned into a parking lot now. The mid-century is definitely reflected by the fins on the vehicles and the white walls and the tire as well

17:08 as the more flamboyant clothing that some of the pedestrians, at least this one who should probably get out of the crosswalk, are in. Nice yellow car. See it in color.

17:19 Were these buildings glass like this before? Like that one right there? Okay, they were, but they were shorter and they were less colorful. So, that's good to see. We still do have birds as well.

17:30 All right, let's take a look at 1985. Synthwave aesthetic. Good. And it's raining. That's actually kind of cool. Okay. And this has a rewind video store here. Now

17:43 the cars are more boxy, which I think kind of makes sense. There were a lot of like wedge looking cars there. We have a city bus, which has some windows in it. Interesting.

17:56 All right. Next up, we have 2005 Digital Dawn. Okay, the buildings are all taller and the facades have changed. That's good to see as well. What do we

18:08 have right here? Bean and wire. I'm not really sure what exactly that is. And the cars have become much smaller. Pedestrians are still there, and the stores have changed a bit, but the

18:20 buildings are significantly taller now. All right, let's check 2025. Okay, we have some bike lanes now in the streets that have appeared. Charge on every corner. Hellion. Okay,

18:38 this was the name of our browser OS. Look, there's like an autonomous delivery robot there driving around. That was cool to see. Do the people have anything in their hands? Sometimes they

18:46 get like mobile phones and things like that. I can't tell, but the street lights are working. And then finally, we have 255. All right, the buildings are even

18:59 taller. And it is like a more synth wave thing. Oh, look at that. That looks like the rumored OpenAI Pebble device that folks have been saying they have. That's actually kind of cool. All right. And

19:10 this person is holding something and has something on their head. So, that is pretty commonly seen. The air is clean. All right. I like that. And kitchen 55. All right. Overall, this was simple. It

19:21 was not as glamorous as I'd hoped to see, truth be told. However, it was competent and the details did reflect the specific era. So, overall, acceptable, but nothing that really blew

19:32 me away. So, next up, we're going to be doing the beautiful watch website test where it has to create a 3D model of a watch, put it in the hero section with a cinematic panning effect. It also needs

19:42 to have an exploded view of the watch. And I have also added into this prompt that there needs to be a special edition watch called the Beijan which shows the which shows the attached image on the

19:53 dial and is an exclusive model. And this of course is just my profile photo from X. So we'll see what we get. All right. So the beautiful watch website has been completed. Let me see how long exactly

20:03 this took. Okay. So right about 64 minutes. I'm very eager to see what it's created for the Bjan. So just take a quick peek. All right. So far, this looks nice. Interesting. We're getting

20:14 some Z fighting on the face of the watch. It also has some issues with the um numeral markers being oddly oriented. However, they are symmetric with one another. So, you know, something I do

20:26 want to actually note real quick is the fabric or the emulated. Actually, that's not fabric. It's like some leather pattern. That's actually one of the more decent ones I think I've seen in doing

20:37 this test. Everything else looks more or less acceptable. The glass is kind of hovering a bit over, but it does have the correct um like hand movement. So, we have our second hand. And I believe

20:48 this is also showing the time that folks in the comments have mentioned gets used to advertise watches because it's the most aesthetically pleasing in that it shows all of the label text and things

20:59 like that. All right, not bad. And it is on a decent looking high-end like circular table. This is the Solstice, of course. Let's go down and see what we get. Okay, that is the

21:12 solstice, which we did see already. Can we move it from in here? Yes, but side to side. If we scroll down, nice. The tide. I like this one better personally. Probably just more of a fan of this

21:23 color. We have some text on there. Same issues with the floating glass and the Z fighting and the numeral markers, but overall not bad. Now, we have two other things that we are very excited to see.

21:33 One being the exploded view. Oh, but the Van is also present. Exclusive upon request. Very good. So, you can't actually just flat out purchase the Van. You basically need to

21:47 be approved as a buyer. And this did a nice job of placing that photo in the back of the um the watch face. So, not bad at all. Simpler in some ways on black alligator. Oh, okay. So, that's

22:02 what the straps are. And then we should also have the exploded view. All right. This seems a bit glitchy. However, there we go. That's kind of what I was hoping to see. Not bad. Can we move it

22:15 around? No, we can't. So, that inhibits our ability to kind of look underneath and see the movement. However, this is more or less a competent result and definitely kind of checks the boxes that

22:26 were laid out in the prompt. However, I'm not overly blown away with it. Okay, there's a nice little I think that was a the spring, but I it's hard to tell. Interesting. Obviously, this one is my

22:39 favorite personally, but you know, this one looks good as well. So, next up, we're going to be creating the Street Yeet game. However, we're using Blender and GDAU. Blender for the assets and

22:49 then GDAU for the game as is listed right there in the prompt. This is a relatively simple prompt and I've just told it to go all out, make it hilariously unrealistic in terms of how

22:58 they get yeated and make it fun. So, we'll see what we get. All right, we're getting some of our street yeet. Like, the game is basically just flashing on screen while it tests it, which is a bit

23:08 of a spoiler, but I have to say it just test yeated an entire building. And that is something that is not always included in this game, keeping in mind this is a pretty simple prompt that I gave it. So,

23:21 that building just got absolutely heated. I'm very, very eager to play this game. All right. Blue cap. It named them.

23:49 This is well done. This is really well. Oh no. Why do they put dogs in this? It's just not right. Dog ye. That's just to the moon. All right. Do you think we can yeet one of these

24:04 skyscrapers? Personally, I don't think so. Oh, yep. Okay, we can. Wow. And all right, let's just eat everything. normally like I don't think I've had one

24:29 before where you can yeet every single thing in this environment. So keep in mind it made all of the assets in Blender by itself and then it used GDAU for the game.

24:48 That was This was quite quite well done, I would say. That was cool. The camera actually moved so the bus didn't get in our way. Good attention to detail.

25:01 Well done. All right. Well, that was definitely something. All right, because that Street Eat game was so fantastic and having it use Blender

25:18 and GDAU definitely took it to the next level, we're going to do it as well with the 80s 3D wrestling game. I'm keeping it extra high on fast right here. And the context window is also set to the

25:28 higher of the two options at 500K. Basically, this is the wrestling prompt that we generally run. I just feel like I want to make it like a little more like and we'll see what we get.

25:44 Well, it seems that our wrestling game is making some progress. All right, our wrestling game has completed. Neon Cage Slamfest 85 is ready. It's a free-for-all neon VHS arena. Pink

25:56 gimmick walk out. Keep hurting people. Pins do not end the night. I'm very excited for this cuz the street eat one was good. All right. The music is

26:11 R for rude. Okay. Oh, wow. Okay. So, we I'm going to turn the music off for now. We have Rex, we have Veil, we have Ivan, and we have

26:25 Set. Actually, kind of interesting the movements there because it would have had to have rigged these. So, let's just go with the largest one here. All right. Oh, wow. That was cool. The walk up was

26:36 actually pretty good. And the ref. I'm going to turn the music back on cuz Oh, okay. All right. The way they're Oh. Oh, no. Get back in there. Why did they get iron curtain?

27:11 Oh, okay. There's I mean D tractor splash. This is hard to play. These are generally difficult to play. Let's just uh Okay, nice. How are we doing? Our health

27:30 is at 32. Sat is the one that's still Okay, Rex just like regenerated. Oh, we just got slammed. Is that a beer keg? It said R for rude. I don't know what that meant in the beginning.

27:51 This is quite chaotic. Get up. Oh, the space bar is toggling the music on and off. I thought it was the jump button. All right.

28:20 Maybe I'll try a different move. Oh, okay. I think we got knocked out. Okay, good. All right, I muted the music there on purpose that time. Okay, I think overall

28:31 this is interesting because it had to have rigged these independently and also the like weird narration over the screen like the dance break where it

28:43 said this is legally binding. This is quite something. So, let's try the it's just chaotic as as anything. And that's I suppose kind of what we wanted. I do believe I actually added that into

28:59 the prompt. Oh, look. Okay, that was kind of cool. This is actually, I think, one of the better ones we've seen. Do you think we can hit the ref? I just want to check

29:08 out of curiosity. Wait, I think we're about to win. Oh, yeah, we can hit the ref. All right. Yeah, this is oddly enjoyable. All right. Next up, we're going to be doing the subway FPS. This

29:19 is the one that is slightly updated because the additional waves after the first one need to be delivered to us via train. So, there should be a subway train that comes in and drops off the

29:29 enemies for wave two and so on, assuming that we can get past wave 1, which we generally can. So, this has to use 3JS because all of our prior references for the specific task also use 3JS. And

29:40 we'll see what we get. All right, the final thing. Our Subway FPS game has been completed. It took around 40 minutes, I believe. It should currently be running at this URL. Okay, so far so

29:51 good. King Street is giving me kind of like London Underground vibes. All right. So far, I like what I'm seeing. Okay.

30:01 We should have multiple weapon types. Yes, we do. Very good. Very good. Are there bullet holes in the environment? Good. Yeah, they are. And they reflect to the

30:11 specific type of weapon. So, as we note, the one that shoots more than one little piece of ammo at the same time has a larger one. All right. Now, we should see the train come into the station to

30:22 deliver the next wave of enemies. Very good. All right. That's what we're seeing right there. Back to the train you go. All right.

30:43 And we actually get to see inside the train. Some folks have always mentioned like, why don't you try to get in it? It the models always put mesh colliders on them. So that's the reason why. All

30:51 right. This is overall, to be honest with you, it's kind of simple, but that doesn't mean it's bad. It just means it's pretty darn simple. There are health kits put down here and things of

31:01 the sort. We also do have enemies that stay there. I wonder. Okay, good. There is a barrier there so we don't accidentally go onto the track. We have a coffee out of order machine. And the

31:11 waves should become iteratively more difficult as we progress. And we can definitely see there are a lot more of them now here. And look at the actual pump on the shotgun. I had not noticed

31:21 that. And the way it reloads is like more in line with reality u considering what reality would refer to in a game like this. All right, overall this is clean. It's not as good as I would have

31:33 hoped to have seen. However, it's clean. I don't think we'll be able to make our way past this one here. Yes, there are significantly more of them. Oh, okay. And just like that, we lost.

31:50 All right, so overall that is probably going to conclude our first look and test of Grock 4.7. This video took quite a bit of time to film. I did also use Cursor for some of the tests as we saw.

32:02 And additionally to that, I had used a Super Gro subscription. I actually ran out of usage partway through because I was on the $30 a month plan. So I just upped it to the Super Grock one, which

32:12 is $300 a month, which is still the most expensive of all the subscription plans for all of the Frontier providers. So, that is something in and of itself. I suppose as a quick results overview, of

32:23 course, the browser OS is one we probably don't need to go over multiple times. The Skate game was interesting because it was competent, but nothing really blew me away. And to be honest

32:33 with you, I think that's probably going to be one of the takeaways I noticed, at least with the games like this or the ones where we had it create something to run in the browser. It was good and it

32:43 was competent, but that was kind of it. it didn't really seem to go above and beyond at any one time. And keeping in mind this was the second iteration of this game where we said kind of just

32:53 give it some more polish and fidelity. I think the same could be said for the city timeline one where this was a competent result. Everything looked good but it was just kind of simple. It

33:03 seemed like it made really good foundations for things but then kind of kept them towards like the foundation and didn't build up to a fully more polished result. The street result was

33:13 very good. That was one where it used Blender for the assets and then GDAU to create the game. That one was a lot of fun and the fact that pretty much every single thing on that map was yeetable is

33:22 not something I've seen before and was a nice welcome surprise. Additionally to that, the wrestling game in retrospect when I actually look back at it, I'd say it was borderline impressive because it

33:33 is difficult to do that, especially when it has to rig the characters that it creates in Blender. Some of the movement, especially I think when we got hit and then kind of like flew back into

33:42 the stands, the actual movement of the player's body was actually somewhat decent, especially considering that with that test, I've never really received a result that's like this is what I will

33:53 compare all other results to. So, and then the final thing I think was the robot arm where it worked so quickly and it was doing pretty well. The problem was it just never quite grasped where

34:04 the car was. No pun intended. It did move it around a bit, but it just was missing like the final 10% there, I guess I would say. So, I think overall takeaway for this model, the big thing

34:14 is the pricing of this model. Now, of course, we ran pretty much everything on fast, which doubles the price, but assuming you don't, it's $2 per million in and $6 per million out. This is

34:24 significantly cheaper than things like Opus, Astra, or forget Astra, but like Soul, and then other models beyond those like Fable and Astra. So, if this is able to actually properly give you even

34:36 like 80% of what those would for a significantly cheaper price, there's definitely a value proposition here. And this can handle a lot of day-to-day workflows. I would go out on a limb and

34:46 say so it is a very competitive model when factoring in the price. Even if it does end up using more tokens for a specific task, I still think it would work out to be cheaper. Although, a lot

34:56 of benchmarks will still probably be coming in for this model. So overall, that is going to conclude our first look and test of Gro 4.7. If you have any questions, please feel free to leave

35:06 them in the comments.