I paid $3,000 for Grok 4. Was it worth it?

Rock 4 just came out and has everyone in the AI space talking. It is the most expensive AI model ever at over $3,000 a year. But here's the thing. Half of the people are saying this is the smartest AI model ever. The other half are saying it can't even do basic tasks. In this video, I'm going to drop the $3,000. We're going to get the most expensive AI model ever, and we're going to see if it's worth your money. Is Gro 4 worth $3,000? I'll tell you this. It'll either be the best money you ever spend or a tremendous waste. Let's find out. So, here it is. Gro four heavy. I'm going to

buy it right now. I'm going to tell you what it is, how it works, and then we're going to give it two very challenging tests. All right, we did it. We bought it. Let's see this. Yep, there it is. Gro 4 heavy. All right, so real quick, what is Gro 4? What makes it different than every other model out there? Well, Elon just dropped this and it destroyed every single benchmark out there. So, if you take a look at this, it is a massive step over Gro 3, which just came out several months ago, and Grock 2, which came out less than a year ago. It's also destroying every other model and benchmarks on basically every metric you can think of. It's at the top of the top. But here's the thing, there's a lot of controversy around benchmarks. You can train AI models to be really good at benchmarks, but in practicality, they're not actually that good. So, the question remains, just because Gro 4 is incredible at benchmarks, does it make it an actually good model? Well, I'm going to give it two tests today. Test one, I'm going to have it build a really advanced 3D simulation. Test two, I'm going to test more of its logic side. We're going to compare those tests with Claude 4 Opus, what I'd consider the other strongest model at the moment. So, let's get into it. This is Gro 4. What is the difference between Gro 4 and Gro 4 Heavy? Gro 4 is their newest model. Gro 4 heavy is their newest model just times three. Every time you give a command to Gro 4 heavy, it's going to run three Gro fours at the same time and then it's going to decide which of the three agents gave the best results and give that to you. So, hypothetically speaking, it should be giving you better results. Is it worth the $3,000? Let's find out. All right, so here's test one

I'm going to give to Gro for Heavy. Then it to Claude for Opus right after to do a comparison so we can know which model is better. We're going to build an advanced 3D simulation world. So, I have this advanced prompt here. I'd invite you to pause the video, copy and paste it from the description down below and put it into your own chat, whether it's chat GPT or whatever you want to use and test this out with me. What we're going to do is build a Minecraft style procedural terrain generator with 3JS. For those who know 3JS, excellent 3D library for JavaScript. It's going to have an infinite terrain that generates as the player moves. is going to have many different biomes. So, forest, water, everything. It's going to be really, really advanced. Daylight cycles, waters at sea level, underground caves, things like that. We're going to see how creative Gro 4 Heavy can get here to see what it can generate. So, let's send the prompt here to Gro 4 Heavy. So, as you can see here, it is now going. What Grock 4 Heavy is doing is it's spun up four Gro 4 agents that are going to complete our task for us. All these agents are going to run in parallel and then the AI will decide which agent gave the best results. So again, supposedly this will give us better results. Obviously, running four agents at once is much more expensive. That's why the $3,000 price tag came with this. So, let's see how it turns out. Okay, it looks like it finished. Uh 10 minutes and 52 seconds in, the Gro 4 World Simulator is complete. That is the longest I have ever seen an AI model work for. 10 almost 11 minutes. That's unbelievable. Let's take this code and let's test it out. So, what I'm going to do is copy this code. Part of the prompt, which I put down below, was to put this all in one file. So, it's very easy to run. So, I'm going to copy the code. We're going to bring it over to cursor. I'm going to show you how to run this. So, what I did was I created a new file in cursor called index. html. I'm going to paste in our file, which is 466 lines. That actually isn't too bad for everything we asked it to do. I'm going to hit save on that, and we should be good to run this. All right, so I open the folder it's in. I'm going to double click, and let's see how this

world generator goes. Click to play. All right, Grock 4 heavy, what do you got here? Let's see what we got. I click. I'm clicking. I'm clicking and nothing's happening. Uh-oh. Okay, let's see if we got any uh errors here. Three is not defined. So, okay. So, we're getting an error. We got an error off the rip. Before we debug this, let's see how Claude 4 Opus did to see if it was able to do anything like this in one shot. It pierce Gro 4 Heavy wasn't able to do this in one shot, which is a little disappointing for a $3,000 model. Let's see how Claw 4 Opus does. All right, so I'm I moved over to Claude 4 Opus. It looks like it's done. I wish it had a timer on here so we knew exactly how long it ran so you can compare, but I would imagine it was uh a little bit quicker. Uh, I'm looking at the preview here. It's all I mean, it's better than a click to start screen. So, that's good. Uh, let's copy the code over and test this out. So, I'm going to hit copy up here. I'm going to go back to cursor. I am going to take all that code. I'm going to paste in Claude's code, which was about 50% more, so 630. Let's try

this now. Let's open this up. Okay, so we are in. Wow, we can move around. Uh, this is an infinite load 3D generator. That's pretty sick. So, I'm moving around here. Is it flawless? No. You can see there's glitches. Uh, but this was a oneshot 3D world it created. There are trees. There are different biomes. Like you can see this is snow. There's grass right here. The lighting changes based on which direction you're facing. That's pretty cool. The water. Okay. Now, okay. It looks like I'm in the water now. This feels like water physics. I'm like jumping up and down in the water. Uh, is this perfect? No. But this is pretty solid for Oh, you can see like the shadows moving as the sun goes around. So, it had the dayight cycle I asked for as well. That's pretty sick. This was all one shot, right? And if I spent another 5 minutes in Clawude telling it what to change, uh, I can probably improve this a good amount and make it playable. Oh, look at the top left. It has the biome, which is snow, and the chunks it's loaded. So, it kind of thought ahead and creatively added different metrics to the top left. I can track. So you can see I can climb up on the snow right here. That's pretty awesome. What happens if I go on the green? Okay, so that's mountain. So it's a biome mountain. So I'm climbing a mountain now. That's pretty sick. Claude did that in one shot. Uh Gro 4 heavy did not. Claude 4 Opus is uh Claude 4 Opus you can get access to for $17 a month, I think, which comes out to uh about $170 a year. Not bad compared to the $3,000 Gro 4 Heavy's in. Um, let's do this. Let's spend just a couple more minutes trying to fix what Gro 4 Heavy did. Uh, if we'll give it another shot to fix what it made. If it can't do it, we'll move on to test two. So, I'm going to go back into Gro 4 Heavy. I'm going to say not loading. Getting this error. So, while Gro 4 Heavy fixes that, just real quick looking at the prompt we originally sent it. If you want to do cool tests like this at home. I like to test the creativity of these models. That's what really excites me about AI models is the creativity part. I don't need an AI model that can answer every question and can solve the SATs, right? Google did a decent job of that before. I like AI models that can be really creative and come up with interesting ideas and build interesting products. So for any test you want to do at home with creativity, I like to use 3JS. So feel free when doing this at home, say, "Hey, I want to build a 3D simulator of this or a 3D simulator that using 3JS, put it into a single index. html file, which will make it really easy for you to run it at home because then all you have to do is go into cursor, create a index. html file, put the code in, and you're good to go. So as Grofor has built this, feel free to run your own tests as well and build cool stuff with 3JS. Let's see how Grock 4 heavy is doing. We refreshed. It looks the same. I'm clicking. Same error. Now we got multiple errors. It can't load an import statement. Not a good sign. Uh Claude Foropus was able to do this in one shot. Grock is now officially in an error hole at the moment. Uh it cannot figure out how to import 3JS, which is frankly something I've been doing with every AI model since Grock 2. I mean, I bet if I did this with Grok 2 right now, it would work in one shot. I built plenty of 3D simulators with Gro 2 uh with no errors at all. So, it seems Gro 4 does not know how to import 3JS uh and is having issues with uh importing other technology like the pointer lock controls. So, we're going to chalk this one up to Claude 4 Opus. Uh it did it significantly faster and it came up with not only great results, at least it had results. Gro 4 Heavy wasn't really able to do this. So, all right, let's move on to the next one. All right, the second

test we're going to do to see if Grock Heavy can handle this and see if it's worth the $3,000 is have it build an interactive stock tracking system. So, the ability to put in stocks, have it put all the data about the stocks on screen in cool 3D ways and see what kind of interesting app it can come up with. I like to do this test from more of the logical side of things when it comes to models because the model has to understand what different indicators for stocks do. It needs to be able to pull in many different tools like stock tracking APIs and visual APIs. So this is a cool test I do with all new models from more of the logical side of things. So we're going to put this in Gro 4 heavy and I'm going to put in Cloud 4 Opus to see which one goes better. So as you can see here the requirements I just sent to Gro 4. It's going to build it all in one single HTML file for us. It's going to use 3JS. So, here are the requirements I gave to Gro 4 heavy. Uh, it has to be in all one single HTML file. So, this is really easy to test. It has to have stock searching, real-time price data, key metric dashboard, interactive price chart visualizations. It's going to use Alpha Vantage, which is a free stock API to pull in the data. Uh, I put this prompt down below as well if you want to run this test with me. You also have a cool stock tracking app after that you can use for yourself. I like to I built my own stock tracking app which is awesome. So that's down below if you want to pause and grab that too. But Gro 4 heavy is gone. It has all four of its agents going. Let's take this now and put this in Cloud 4 Opus to see what we get as well. So I pasted that in Claude 4. We're going to let that run. Claude 4 seems to run a lot faster just cuz it doesn't send five agents at the same time to work through it. So far it doesn't seem like it's that much of a disadvantage for it. It was able to oneshot an entire 3D world while all four agents of Gro 4 Heavy were not able to do that. All right. Uh it looks like Grock 4 Heavy has uh no response. All four agents basically said, "F you screw off. " Okay. Uh I guess we'll just run it again. So I put the prompt back in. Grock for Heavy is going to go at it. Hopefully the agents want to do something for us this time and uh don't tell us to screw off. So we'll see how that goes. over on the claude for opus side. It looks like they are done here. Let's take this code and throw it in cursor and see what we can do. So, I'm going to hit copy here. Moved it into cursor. I hit save. I'm going to get my free API key from Alpha Vantage, which is just a stock API. You can do this if you're following along with me. I refresh my screen and it looks like

the app is working. Let's do this. Let's enter in our API key here. I'm going to delete the API key after. So, don't even bother trying to copy that. I'm going to hit search. Let's see if we can get Okay, it looks like a chart just loaded here. Uh, let's see. Can I scroll down? Okay, I can scroll down. Uh, the chart, I'm going to be honest with you, looks terrible. Oh, look at this. The site is kind of buggy. Uh, it scrolls really slowly, but it does look like it's getting real time data. And if I scroll down, I have a ton of different pieces of data I can do here. It's scrolling back up. I think it's trying to add different animations or something which is causing this to slow down. But either way, you can see all the data around the stock here. And it even has a stock chart which let's see what happens if I do. Oh, if I make it one week or one month, it starts adding actually the technical data on it. So, the SMA, the price, it is a little bit glitchy. It keeps going in and out. I think it's probably around the fact that I gave it like 20 different animations I wanted it to do. Uh, but if I probably spend a few more minutes working on this, I could probably iron it out and make it look good. So, is this as slam of a dunk as the Minecraft simulator did before? Not quite. But this is still pretty impressive for a oneshot prompt on a $17 a month AI model. Let's go back into Gro for Heavy. We sent the prompt for a second time. Now we're 5 minutes and 18 seconds in. Oh, completed. Okay, so it looks like those meters that measure how long it's going are not exactly accurate. I did it about a quarter of the way through. Uh let's take this code and test it out. So I'm going to hit copy on this code. I'm going to go back into cursor. I'm going to paste this in. It is significantly shorter code uh than Claude. It's only 369 lines while Claude was I think over,200. All right, maybe it's a better, more efficient coder. Let's see. I hit save on this. Let's refresh here and see what the Gro 4 heavy $3,000 code looks like. Oh, okay. So, let's see what we got here. It actually works this time. Um, which is a that's a win the fact that it works. Uh, error. The demo API key is for demo purposes only. Please claim your API key. Oh, we have the API key. Okay. So, where do we put that? It's not exactly clear where the API key goes, but let's search. Uh, nothing quite happens when we search here. Does Grock 4 give us an explanation of where to put the API? No, there's no quite Grock doesn't really give you an explanation. and it just kind of spat out the code and absolutely nothing else. Claude's strength has always been the fact that it's very humanlike and warm and it feels like you're interacting with the person. That's why Claude's always been kind of my go-to model. Like if you look at Claude, it gives you an explanation how the code works and how to implement it, which is always nice. That's always been my favorite thing about Claus. It always feels the warmest and the most humanlike out of all the models. Grock 4 is a little cold, right? It doesn't. It just gives you, hey, here buddy, take the code. Doesn't really give you an explanation. While Claude walks you through what to do. So, it's not exactly telling you where to put the API key. I I guess if we want to find the API key, I can search in here. Let's see. API. Okay. So, it has a spot for the API key. Let's put that in. Paste that in. Save that. Go back to our Gro 4 heavy. Uh, and let's see how this goes. I'll search IBM. Oh, wow. Okay. So, I refresh

it. Instantly has the chart open. So, it did fix the issue. Oh, wow. Look at this. This is a good looking chart. I hover over. Okay, it has every price. You can see this is max 2. So, this is all the n since 1999. Let me go back 3 months. Wow. So, it has volume, it has PE ratio, it has everything involved. This is sick. It has the SMA like we asked for the total market cap. Let's see what happens. We add to watch list. Okay, so it has the watch list here. Let's just try one more stock. We'll put in Apple. See what we get. Okay, so Apple loads and again the really nice chart down here. I really like the way this looks and it includes all the data we asked for as well. I do not think the clawed one had volume. This is really nice. Now, there's a couple things I

want to say about Gro. I've been testing Gro 4, the regular version, since it came out about a week ago, and I've been having experiences similar to what some other people have been having, which is actually pretty in line with what we just saw with Gro 4 Heavy, which is this. At Gro 4's best, it is the smartest model out there. But unfortunately, Gro 4 isn't at its best all the time. In fact, it's not even at its best most of the time. It's kind of hit or miss. Like what you saw with this stock tracking portfolio, like this is actually really well done and really advanced. And it took two prompts, but even for two prompts, this is pretty incredible what it was able to come up with. At its I would say this was better than what Claude spat out. I really like this. It's not slow at all. It's not buggy. It's very usable. It's beautiful. I really like the taste it has in there. And it has everything we asked it to do. So, at its best, it's incredible. The issue is at its worst, it just doesn't work. The issue we saw with the 3D world simulator is actually very similar to the results I was getting just using regular Gro 4 is a lot of the tasks I'd give it. It just would mess up. It really wouldn't understand what I'm asking it. It would disagree with things I'm saying and it would just get into a rabbit hole very quickly. Here's the thing though. If you pay attention to Elon's X, he has been announcing new features to Grock the last few days since Gro 4 came out. So, they are continuously improving. It's not the type of thing where they just release it and it's done. They have added new features. They have improved it. So, I do believe when they iron out Gro 4 and Gro 4 heavy, it will be an awesome experience and it might actually be worth the $3,000. But at the moment, just based on my tests, it's been very unpredictable. I cannot recommend just yet. Gro 4 heavy. I do believe the regular Gro 4 at the $30 a month price point is worth experimenting with. Their voice mode is excellent. But if you're a power user and you're trying to build big applications, I would wait on Gro 4 heavy for now until there's some improved. for the $17 a month that the beginning version of Claude 4 is, you get Claude 4 Opus, which is able to build amazing applications very quickly. I'd say right now if you're doing more logical tests like physics-based questions, math-based questions, things like that, I go with Gro 4. If you're looking to build applications, if you want a more humanto human creative connection, things that require more human type creativity, I am leaning more towards Claude. I hope that was helpful. I hope the $3,000 becomes worth it very soon and they release some more improvements to make it more predictable. If you learned anything at all, make sure to hit subscribe, turn on notifications. All I do is make banger AI video educational content. Leave a like if you learned anything, too. and I'll see you in the next video.