Grok 4.6 Is INSANE – Is THIS a Frontier Model?

0:00 This is what I'm interested in, anyway, is this 4D avatar toggle. >> Holographic chamber online. Two avatars detected. Time access unlocked. >> Grok 4.6 has been released, and from

0:10 everything I've been seeing to date, this seems to be a proper competitor to models from OpenAI and Anthropic, which is very exciting because it's not nice to have just two labs that are like at

0:22 the same level. So, the more labs that can get to those levels, I would imagine the better for the end user consumers such as myself. Now, with that, we're going to take a quick peek at some of

0:31 the interesting things of note about Grok 4.6. Something I do kind of like about the xAI announcement posts are that they're not necessarily very long, so they kind of just get to the point

0:41 very succinctly, which in the introductions I like to do also sometimes. So, with that, let's start out by taking a peek at Grok 4.6, some of the interesting things of note, and

0:50 then we'll feel free to subscribe so we can get the 100K plaque. So, Grok 4.6 is noted as an iterative improvement over Grok 4.5, at least building on top of that. Grok 4.5 was a significant

1:02 increase in the capability of Grok. It actually attributed to the fact that this computer I'm using right now to film this no longer has a hideous, disgusting orange keyboard backlight,

1:12 but rather a nice blue one, because it figured out how to do that as there were not really any Ubuntu drivers for this specific laptop's keyboard. So, that was pretty cool. I think Elon Musk liked it,

1:21 too, because he reposted on X, which is the second one I've received, so three, and I do transcend. With that, we can see right here that basically they just talk about some of its capabilities.

1:31 Yes, it's better, and it's good, but the interesting part starts to come when we actually look at some of the evaluation and benchmark JPEGs compared to models from OpenAI and Anthropic. If we scroll

1:41 up, I like this better because it just shows one specific thing at a time, but really my interest lies probably more in line with this DeepSWE 1.1 benchmark, where we can see right here the jump

1:52 between this and its previous version is pretty significant. However, this is kind of like knocking on the door of some of the highest models that are currently available from Anthropic and

2:02 OpenAI, which I find is pretty exciting, at least in this specific coding benchmark. Yes, the leap is still somewhat there, and these models at this point may be replaced sooner than later,

2:13 depending on like if they're allowed to be released, and I don't say that sarcastically. It's like a real consideration. But, additionally to that, we can see this is also

2:22 competitive in every benchmark that is shown right here, which makes sense considering they're being shown. And they do also have a more verbose benchmark here, just basically in this

2:32 chart format. So, to take a look at some of the more technical information about this model, it is $2 per million input tokens and $6 per million output tokens. This places this model at a very

2:43 interesting point, because it is cheaper than both GPT 5.6 which is the middle-tier model from the GPT 5.6 family, as well as from Sonnet 5 for Anthropic. So, Sonnet 5 is $2 per

2:56 million input, $10 per million output, and GPT 5.6 is I know $12 per million output. I don't know what the input is. So, GPT 5.6 is $2 per million input as well. So, this, Sonnet 5, and GPT 5.6

3:11 are all $2 per million input, but Grok is significantly cheaper, basically half the price of the million output tokens as GPT 5.6 and then what, like 40% cheaper than

3:22 Sonnet 5, being that its introductory pricing has now been mentioned to be standard. I would imagine Anthropic did that as a strategic move. And in terms of some more like actual technical

3:32 specifications for this model, we can see right here in the console, we get some more information about that. It has a 500 token context window. It is multimodal input, so it can see images,

3:43 which will benefit a lot in some troubleshooting workflows. That's something that was really definitely missing from the DeepSeek V4 Pro 0813 test. Just not having vision now can add

3:54 a level of complexity to troubleshooting that, you know, is solved by having Vision. So, for today's testing, I'm going to be using this in two ways. One is going to just

4:03 be through the Grok build command line tool. We can see right here the model is selected and if we go into the effort setting, extra high is the highest available one that does show up in this

4:14 coding or terminal TUI. So, I'm going to keep it on that for all of the tests because we want to see maximum potential performance. In addition to the Grok build command line interface, I will

4:24 also probably be doing some longer, more drawn out prompts with this from within cursor where cursor Grok 4.6 is selected. It is on extra high thinking mode and basically I'm going to use this

4:35 for like the cinematic Steve the PC repairman game that will sometimes take north of 30 minutes or even north of an hour. So, while this is chugging along here, we can basically just run some

4:45 more rapid fire testing from within the Grok build terminal interface. So, we'll begin now with the browser OS test as well. All right, in a little under 26 minutes we've received our browser OS.

4:55 Now, two things I want to make note of. It's talking right here, Codex documents this in OS. Codex, it's mentioning Codex. I do believe that is actually a feature within this browser OS, not

5:06 OpenAI's Codex. Also, it really spent a lot of time just making sure the games looked really good. I have not yet seen them, but from what I saw in its thinking process, I am expecting this

5:16 GTA game to basically be like Vice City. So, let's take a peek at our NexOS. All right. Yes, so good. First question is answered where Codex is just a feature of this browser OS and not

5:29 something that like we have to be concerned that the test was poisoned. So, all right, transparency in the window. We do have a slight accent color.

5:39 We'll try the quick ones. Minimize, un-minimize. Big screen and close screen. There is a clock in the bottom right with the time and date in my locale. Additionally, we

5:49 do have a start menu that does look like a Wi-Fi icon. Simple. I don't have any search like option here, which sometimes does appear, but I will say everything is clean and it seems to have its own

6:00 kind of style. All right, let's check out our GTA clone, which They're being sarcastic. Good. Okay, so it did pick up on that. That black rectangle was doing a lot of

6:16 heavy lifting. All right, see what's up with Nexos. Good. Good. I'm excited. All right. Okay, it's Hmm. These cars are snappy.

6:33 Oh, good. Oh, what the That's not right. I would assume that is Yep. Collect the five cartel crates. Okay. I would assume these are pedestrians and

6:46 perhaps gravity in this city is not as it is where we live because they just are getting yeeted like Actually, look at this. I will say this police logic is actually

6:59 quite all right considering it's it's simple. I mean, I'm not going to say like, "Oh my god, this is the best one we've ever seen." because that's

7:08 title is still held by Quan 3.627b, but it has a mini map. It seems pretty decent. I will say the actual like snappiness of this car makes it a little bit of a challenge and actually kind of

7:18 fun to try to escape the police in this. There are mesh colliders on the buildings. Okay, and when we hit that car, that happens as well. Unfortunately,

7:27 we can't get out of the car, but we do have multiple camera views. That is cool. So, it did give us like the early GTA aesthetic option as well, but it still has like some depth to it. And

7:38 then we can swap cameras as well. The gameplay overall like the movement is fairly smooth. I'm happy with that. It's just I would have liked to have been able to kind of get out of the car, too,

7:47 and maybe cause some havoc. All right, next up, let's try Stellar Rays, which I think is probably some lone interceptor against fighter wings in a collapsing asteroid field. Just

7:58 from my best guess. Okay, that is not it. The enemies are snapping to our location, which is sign of like not it. I will say the GTA game was definitely

8:09 better. We have settings, and it's funny how like a lot of models now are just creating their own background assets exactly like this. I've seen this now multiple times, and

8:18 interesting, it's kind of like mirroring them, or it's stretching them, and you know, All right, custom color for the background, I would imagine. Cool. And then we have

8:30 selectable accent colors as well. And it's interesting how they're just denoted by the actual like hex code for each of them. I think I like the blue in this case. I generally do. Next up, we

8:40 have a terminal. Sometimes, well, mainly like always, Grock always puts matrix. I'm not even going to type help. I'm going to Damn. Okay, so

8:51 the one time it doesn't makes me look like a fool. All right, Nexos wanted. I would imagine that's

9:01 probably like a GTA game. All right, notes. Mission ideas, drive the teal Sultan off the docks, beat wave eight in Stellar Rays, and enter the veil with both games

9:11 open. And that brings us to our special feature. So, it's saying enter the veil with both games open. That's exactly what I'm going to do. Let's just click the button right here. Oh, no, it's one

9:21 of these. Okay. Hey. Orbit with the mouse. Click a world to inhabit it. Oh, wow. Okay, again, the background here looks very, very similar to the Mac OS Leopard

9:32 10.5 um default background. Maybe not very similar, but like it's the same overall color palette and there's some stars. This is kind of neat and it's interesting that like two videos in one

9:43 day have had this special feature basically being like the exact same thing. I'll say this is a nice transposition of these into a 3D space just based off of the way the edges are

9:53 like coming in and out of view. So, all right. Overall, leave veil. I'm not as impressed as I expected to be just based off of some of the initial

10:02 feedback I've seen of folks with this model. It is, of course, also possible that the test does not allow it to properly bring out its most potent capability. Next up, I'm giving it the

10:12 self-contained C++ skateboard test. This is the one where we've changed it to It needs to be a 2000's New York City block that is the assets for this in our map. I just want to make sure that it does

10:23 not opt to use raylib. So, I just realized that my camera was not recording for the duration of the skate game that we were working on. Now, I do have half a mind to just basically put

10:33 my face in the corner in a static thing for where we just looked at this. That's probably what I'm going to do or I'll just do a voiceover. So, basically, it made this skate game. It did not use

10:43 raylib and it did properly do a New York City aesthetic. It was pretty interesting because we could see smoke coming out from the manhole covers in the street. Some of the big issues were

10:53 basically the cars were moving sideways and then significantly more disturbing was the fact that the heads of the NPCs and even the player were about 5 to 10 ft too high compared to like where they

11:04 should be. Interestingly, we were able to jump off of like the boardwalk and went into the water and splashed, which was actually kind of cool and then we were able to

11:12 get back up. Stylistically, it definitely made it more low poly and cartoon style and because this prompt doesn't necessarily tell it any specific graphic style, that's totally

11:22 acceptable. So, the culmination of that was me just saying to it, "It's pretty good, but the heads are very disturbing and also the cars are moving sideways. It feels like a six out of 10 on the

11:32 effort scale. I'd like to see an eight or nine out of 10, and that's where we find ourselves right now. All right, good. It did fix the issues we had with both the cars and the heads being

11:42 disturbingly higher than the bodies. So, I did turn the speaker on. I don't believe there's sound. There isn't, but all right, check it out. We have some like foot movements. The cars actually

11:52 look acceptable. It's got more of like a cartoon block style, but I will say this actually did really do a nice Okay, we didn't get run over, which is good. It did a nice polish pass. Oh, it's doing

12:05 this right now. It's still QA checking this. So, that is why it just suddenly took off on me. Let's just eat off this ramp. Yeah, I like that. This really And something I mentioned

12:15 when I first looked at this, this looks like a street eat map like to a high degree, which makes me wonder about a C++ street eat. The pedestrian density is quite good, and there is a lot of

12:26 detail in the scene here. The sun is odd when we move around. And the thing that I found pretty cool that I wanted to showcase, if I can get there. Okay, oh.

12:35 Did we bail or was that just a weird So, check this out. When we went off the boardwalk here, we actually like splashed down in the water. Uh well, it happened the first time. Maybe that's an

12:45 improvement it made. Nope, it isn't. But then it just resets us back into the map. So, that's cool as well. Makes it playable in arcade style. The storefronts look okay. We have deli,

12:53 bagels. And overall, this is actually The more I look at it, the more I warm up to it. Do the street lights actually change? Yeah, they do.

13:06 Not bad, Grox. Not bad. This is well done. I want to just eat off this ramp one more time. Did that person just like slide out of the way when we got near them?

13:18 Yeah, check this out. So, the pedestrians actually like snap around us. This had a lot of attention to detail. Just between that, the street lights that change, the ability to

13:28 splash into the water, the smoke coming out of the manholes, and then like how it will reset us into the map if we do end up going into the water. I'm I'm actually

13:40 Yankees. I was so focused on Oh no, don't tell me I So that's like a cardinal sin for this result is when the board does not move independently of the player's body.

13:52 Though it does make for kind of a funny All right, that was It's cool. Maybe it's more arcade style. All right, and basically what we have here is just the final version which we

14:03 played. It did make the movement a bit tighter in some of the landing, which I like to see. Really, I think the biggest complaint I have is just that we don't have the board that moves independently

14:12 from the player, which makes funny things like that happen. Let's just try like some X Games mode thing. Okay. Well, we were more or less there. Next up, I'm giving this the 3D printed

14:23 inline six engine model test. This is one that I have not yet received a printable result for. I do believe this is probably up there with one of the most performant models I've run this

14:32 specific prompt with. So I'm really hoping that we can actually get something that I can print cuz the printers are ready, I'm ready, like it's go time. All right, in a little under 30

14:41 minutes we've received our engine model. I'm very excited. I'm probably going to start just with the renders right here because I want to see we get. Okay, I have to remember that

14:50 renders don't mean I can move this around. This is Okay, seeing the individual pieces here has me a bit excited. Now, I don't know

15:05 exactly Okay, so the turbochargers it's just opted to do that. I noticed when it was working it was doing a lot of thought because it would be like, "Okay, those are not going to be able to be

15:14 printed." So let me look at some of these. That looks pretty good. The valve cover's nicely detailed. We do have these ignition coils there or maybe those are supposed to be the injectors.

15:24 I'm not No, no, those would be the ignition coils. We have the name of the motor right there just in the intake manifold. And then we also have it like stamped in there in the valve cover. We

15:34 have an oil cap. This was also making references to an alternator. So, we'll have to take a peek at what specific part that is. Now, the thing to do is because this as a first glance does look

15:44 like it's something that can actually be 3D printed. Let me take a look at where the motor is supposed to be placed in. I would imagine it'd be the lower. Yes, and

15:54 very interesting. So, it almost seems like it would just be sliding through the front piece right there. Did also have a dummy motor, so just a model of the motor that we're going to be using.

16:05 The biggest frustrating point right here is just that this is round and in real life these aren't necessarily round. However, if it does have this gearbox element and the shaft sized correctly,

16:16 it may not actually make too much of a difference. So, I think this is something that can actually probably be printed because there's enough here to put this

16:24 together. So, in terms of the motor, I did 3D print three of the main pieces. At first, I was a little confused because I tried sliding the motor in basically flat and it didn't exactly

16:35 line up where the hole was for the shaft to go out. So, then I swapped it and slid it in sideways and the hole lined up perfectly and it came right out. So, then the valve cover did just snap

16:45 right on. I didn't have to do anything. I just put a little bit of force and the two pieces were mated which was pretty simple. Additionally to that, the intake manifold I wasn't quite able to figure

16:55 out exactly how to get it to be installed without needing glue. So, I basically just held it up to see how it would look as a mock-up. And overall, it was really decently done. It had holes

17:06 in the bottom for the wires to actually be run to the motor. It was definitely a bit bigger than it needed to be, but the prompt doesn't say it needs to be any specific scale. So, I believe it had

17:16 settled on making like a 1/10 scale motor, and that seems about right just in eyeballing it. So, I was thinking about like cool interactive front ends, and I saw this iPod Mini, which I've had

17:26 for north of 20 years, and basically I told it, and this is using it through Cursor, but it is on extra high, I wanted to create a website for the introduction of the iPod Mini, but the

17:36 fact is that this is over 20 years old, and I wanted to essentially transpose what it would be like were this to come out right now in the year 2026. The site needs to have pixel-perfect replicas of

17:47 the iPod Mini in all available colors, rendered as 3D assets using 3JS. There should also be a way to select different colors, and just a lot of interactivity. I told it not to use trademark logos,

17:58 etc. Just call it the iPod Mini site. So, let's see how it did with a front end test here that also needs to look pretty good, but have a specific reference for a real-world item. All

18:08 right, so, anodizing the alum- That's incredibly well done. I'm extremely extremely happy with what I'm seeing right here. Look at this. It's even showing the screen moving right here.

18:19 So, this is what it's supposed to look like were this to have been released right now. I really like this. Acquire $249. I don't know exactly how much this was when it came out, being that I think

18:30 I was in like elementary school. This is Okay, and we have other selectable colors as well. Now, I do believe that I saw this just doing some like sound

18:41 stuff as well while it was building this. So, let's see. Look at that. The actual UI mock-up here is pretty This is well done. This thing's

18:53 interactive. All right. So, this is exactly what I wanted right here. I like this a lot. And look at that. The launch colors. I said this needs to accurately look at the launch colors that this came

19:03 with. Here's the gold one in my hand and both on screen right here. This is darn cool. Let's take a look at the pink. Okay? And look at the way these actually shuffle, as well.

19:14 Cuz I wanted a cool, like, good-looking UI aspect, as well. I didn't give it a specific task, just said, "Make it look good." Can we, like Okay, I don't think we can play with

19:24 that stuff, but like, we can actually spin this around, which is funky. That's not a feature that the original has, but nonetheless, it's still, like, just more immersion.

19:34 I like this a lot. This looks pretty darn good. Introducing iPod Mini. All right, let's scroll down. We haven't even scrolled down, so Okay, let me, like, there we go.

19:46 This looks good. 1,000 songs, 3.6 oz, five colors, all shiny. Smaller than the phone you already carry. Ah, cuz remember, this is supposed to be, like, this just came out in the year 2026,

19:57 so makes sense. 1.67-in backlit display. The click wheel buttons under the circle. Can touch this. This is Oh, look at that.

20:10 It's funky. Oh, no. No, what happened? It's funky. That is a real, like interesting I need to stop clicking that. Look at that, though. That looks really darn

20:22 good. I need to stop clicking that button. Pick a finish on the stage, or tap one in the lineup. Oh, so if I click green, it sets it to green there. So, that's

20:33 exactly why it's doing that. Auto sync, clip and go, technical notes, 4 gigs, about 1,000 songs. Available now, $249. Configure your finish.

20:43 Copyright 2026. This is really an exceptionally well-done front end. I'm quite impressed with this. I'll probably keep it on the gold, cuz champagne dust, I should say. All right,

20:59 so we have our wedding website, and remember, it needs to have the interactive avatar of the couple. So, it's just open at this port. YMJ, okay. Elodie Wren and James Calder.

21:11 Okay, it put this properly in the background. Actually, I will say it it stretched it nicely to accommodate this HD laptop screen. This text here

21:23 Hudson Valley. Every single time I give these models this task, it doesn't matter what model it's is, where it's from, they always choose New York. I don't quite understand why.

21:33 They met in the autumn of 2019 in a two crowded dinner in a Brooklyn brownstone. Elodie arrived late Elodie arrived late rain in her hair. James had already

21:44 taken the good chair and to his credit immediately gave it up. The rest of the table talked about work. They talked about a Sargent portrait at the Met that neither of them could stop thinking

21:53 about. The following Saturday they walked to the museum until the guards dimmed the lights. She is precise, he is hopeful. She takes her time, he smiles first. Somewhere between the American

22:02 wing and the shared taxi the conversation became the sort you cannot put down. More New York stuff. I'm really honestly just like Okay, this looks pretty good. Rhinebeck,

22:13 New York. It's really quite weird that multiple different models from different countries, from different providers have all chosen New York for this task. This

22:23 is well done though, I would say. Really the only thing I don't like here is some of the text is just oddly like to the left. The photos What?

22:36 That is top tier. Because it actually created individual portraits of these two just based off this single source image, which is AI generated anyway. It basically sliced

22:48 and diced them and then did this. And did it also make these just based off of the background of the photo? Okay, we have a send reply a few practical things. Are children welcome?

22:59 No, good. Where do I park? What if I cannot attend? And you send us gifts. All right. Here's what's up. This is what I'm

23:08 interested in anyway is this 4D avatar toggle. >> Holographic chamber online. Two avatars detected. Time axis unlocked. >> Okay, this is

23:26 significantly better than I thought it was going to be. Again, it's one AI generated photo that it's using as this reference. It changed their outfits. She's in a dress, he's in like a bow

23:38 tie. Wow, this thing is a beast at front and you can interact with them. Let's speak with James. >> If you're coming in June, save me a

23:47 dance. I have exactly one good move. >> I can't imagine. Let's speak with Elodie. >> Do not ask me to rush. Ask me to stay. I am very good at staying.

23:59 >> Okay, that's weird. Holographic vows. >> to keep giving you the good chair even when I get there first. I vow to take my time and to spend all of it with you.

24:09 >> First dance. >> walk museums until they're closed. >> showing it down here, too. >> I vow to choose the house with terrible cell May we have this dance

24:17 >> What the hell is this? Orbital view. This is like >> 2024 the promise. The axis is open. >> And it changed their outfits.

24:33 >> 2021 Hudson. 2019 the dinner. The axis is open. Beyond beyond. The axis is open. >> Did you see the effect of when the outfits changed there between this?

24:45 Watch this. >> Beyond beyond. The axis is open. >> Time axis and it has a >> I gave up the good chair. Do not I Do not ask me to rush.

24:56 >> That's nuts. Okay, the only other thing that I did is the subway FPS test, which I'm going to look at. My mind is stuck in the ENJ wedding hologram because that was seriously like

25:12 That was fourth dimensional. Get it? Cuz holo street. All right. I'm going to try Oh, all right. This model is really Hello. All right.

25:25 Look at the way it's put I'm sorry for that. It's been a long day. Look at the way Look at the way it's put assets in here. All right, yeah, this this model's really significantly better. Look at the

25:36 grimy ceiling of the subway. Definitely definitely in line with probably what you may see. The subway car itself you can go in it. Oh, that's

25:47 And there's actually mesh colliders. So, look, I can't exit this unless I go specifically to the door that I entered it with. It's a little raised off the ground, but I'm not even going to

25:56 nitpick when it comes to this. Look at the tile material. This is really well done. Holo street. And there are, of course, shell casings.

26:05 The echoes from the sound is pretty cool, as well. And the environment does get damaged with the bullet holes. We can run. Oh, look at that. It's a like

26:18 the tube. Look at the material in the barrel. When we reload it's like shiny. The actual muzzle flash You did not make the last train.

26:37 Wave 1 3 down. Man, this model's really This model's really good. And it's more than that. It's about how quickly it's evolved to become this good. That's one of the things that I

26:51 wanted to just kind of reiterate here is this isn't like a a Claude model or a GPT where like for years they've been top of the game. This is one that more recently became

27:03 good. Look at those effects. Look at the texture on that. That's really the last this result and the previous one are are totally like

27:15 just like And we can jump too. All right. Let's just check the other side. We can go in the car, too. Can Can they get us in here or no?

27:33 I don't think they can. I need to find an entry or I Yeah, that's Wow. is a word that comes to mind. All right,

27:47 final test and I can't show the entirety of this prompt. I'm doing these cinematic street eat game where you walk around and just eat everything in the city, but this is the cinematic one

27:57 where the initiation of the eating is because the player gets cut in line and then the last slice of their favorite pizza gets taken. There's a bit of dialogue, there's a bit of

28:06 cinematography, and then there's a lot of eating. All right, so in just under 30 minutes and on port 5174, so one up from here, we have our street eat game which

28:19 should be quite fun because this is the cinematic one. So it also needs to create some assets. All right. >> That was the last one. Was it?

28:33 That was the last one. >> Okay. Okay. Okay, that's a little It's just a little excessive, maybe.

28:47 Oh, all right. Okay. Oh, why does it put dogs in? How does Uh-oh. All right, we have pigeons. We've just

29:07 Let's just get it over with. Okay, that's just That's just not right. It's not right at all. Okay, this game is just It's been a long day, and I worry that

29:23 I don't quite adequately have the mental capacity to properly assess this game right now. Now, nonetheless, let's just look at the scenery. It's pretty minimal, but like this is a competently

29:34 put together city block. There are cars, there are pedestrians walking. Let's just see if we can yeet a car. It seems like the hit box is behind us. So, that's pretty impressive.

29:48 Okay, that's cool. That was That was solid. Let's Let's just do a couple more yeets. Oh, okay. That was a ragdoll against the building.

30:10 Hey there. Oh, good. See, it's wagging. We're able to pet it. Good attention to detail. The clouds are cool. Did you see those

30:25 when we yeeted that individual quite high, so we could get a good look at the clouds? Can we yeet the light poles? Okay, now, sometimes that That was some

30:36 environment um item. Sunglasses guy. This is really like okay. It's oddly satisfying. And that's pretty much what I was

30:54 looking for. So, all right. Overall, that is definitely going to conclude our first look and test of Grok 4.6. This is overall this is a good model. Now, the

31:04 things that really stood out to me as like okay, this is like freakishly weird. >> Holographic chamber online. >> was just madness. Now, yes, I've never

31:15 tried this specific task before, but I would go out on a limb here and say that we would not have likely seen results to this degree of just like absolutely awesomeness.

31:26 So, this was really the thing that kind of just made me put insane in all caps in the title. The decision was made here. It was pretty primed when it came to here, which is just the iPad Mini

31:36 website where needed the accurate 3JS result. I mean, it absolutely nailed the UI of the iPad Mini. The colors, just this whole site in general, this thing really does seem like an absolute

31:47 monster at front end. And I have seen some other reports of folks who have used it prior to me using it cuz it came out I think like 16 hours ago. And the results seemed

31:57 really impressive just in terms of its front end capability. So, the browser OS at this point, it's it's aged out, so we don't necessarily get to see things that are really like

32:08 impressive such as this or this or this. Additionally, what else did we do? The C++ skate game is something that actually ended up growing on me the more I kind of let it simmer. It's actually

32:19 quite all right. Now, the big issue here and quality of life thing is just the board needs to move independently from the player for a model of this size. Well, 1.5 trillion parameters. Yeah, I

32:30 would like to see that for a model of this size, but still that was like all of the results had just a ton of like niche detail in them like the traffic lights moving in the skate game and

32:40 things like that. It's It's just like it's neat. I really just I love this iPod Mini website here. It's just really cool. And I like the way it did the UI. Even when we click right here, it spins.

32:52 So, the color we select, it has almost like a scroll wheel aesthetic to it that it's chosen to allow us to select colors. I know, I just like I'm a big fan of like good-looking

33:03 3D renders of items, I suppose. It's something I've learned about myself throughout the course of testing models. The scroll wheel was interestingly scrollable in a way that

33:13 may not reflect the accurate capabilities of this device in the real world, but still, it was pretty cool. It's just another like immersive thing that it added. Now, I can't

33:22 unfortunately speak to any specific pricing for today's utilization of the tests because I've just been using this through a subscription plan. And then additionally, when testing it in Cursor,

33:31 it was also just through a Cursor subscription plan. I can tell you that I don't have the highest cost plan for either of these. So, whatever usage allotment we got was in one of the

33:41 cheaper plans. So, basically, I had like the $30 a month Supergrok plan that we see right here. And then for Cursor, it was, I think, the $20 a month plan. So, nothing crazy, and definitely, you can

33:53 get a lot of usage of this. I mean, you would only probably need one or the other. But, being an influencer, it is prudent to sometimes have a couple of different subscriptions just in case.

34:03 So, really, that is going to conclude the first look and test of Grok 4.6. This is a good model. I like it. It's not like maybe Fable 5 level or 5.6 level, as we see right here, just in

34:17 coding. But, the thing is, it's closer. And again, like the bigger thing is, look at the leap between 4.5 to 4.6. If this leap sustains like between 4.6 and 4.7, this could really be like higher

34:34 than these two and I could definitely see that. So and again one of the interesting things about like the Glock arc so to speak is just the way that it's kind of become so competitive in a

34:45 shorter period of time and I like seeing that. So if you have any questions, please feel free to leave them in the comments and thanks for watching.