Grok 4.5 is Back in the Game!

Grock, who has Oh god, we've just opened up a whole new can of worms, isn't it? What is Grock? Command AI. Grock is XAI's model who worked with Cursor on the new Grock 4.5 model. And that's actually apparently not a bad model at all. So, we have Grock 4.5 compared to the other best models and it's respectfully good and it it is a surprise. It is a surprise especially because they were so much behind. They were a joke, you know, they were a meme, but now it seems like they are catching up. Uh, everybody is catching up and this begs the question, how are they going to win? But what's interesting is that they have a lot of um infrastructure and they own cursor and this might have have to do something with the data that they have from cursor uh which will make them even more and more powerful with the time. But either way, we have a new player in town. Not not really new, but they're back in town. They're not really going for intelligence here. They're they're kind of going for speed and cost which kind of makes sense because you got composer which is cursor's model. Um but it is their smartest model obviously and trained alongside cursor. So this is like what we see with the benchmarks with the deep speed one which is the closed source uh you know arguably more reliable one. You got fable max at 70 GBT 5.5. Be interested to know where 5.6 sits there. Uh but just below in between Opus 4.8 and GLM 5.2. Again, GLM, you know, mingling amongst the the the the big boys there. Uh, sits right there. And again, it's a it's a pricing thing with regards to I mean, this was apparently one prompt, this little 3D thing, which is quite cool. Yeah, it's Yeah, it's pretty cool. Uh, f like 4.2 4.2 times fewer tokens than Opus Max. um 80 tokens per second which I didn't think that was that great but apparently so I was getting 40 tokens per second on local AI and I thought that was I mean it was good but like but this is smart and doing 80 m yeah so 2 million in and 6 million out so pretty good um yeah I think it's I think it's now finally a bit of a contender you can actually use it for free using their tool Gro build uh which you can you know install with this here. I've actually got it installed and it's quite a nice little quite a nice little interface to be honest. What's funny is that they've done what they've done is you can actually type in agent to start it which I thought that was quite cheeky. Um and nobody else uses agent. Exactly. Exactly. That's what I'm thinking. Obviously, you can type Grock as well, but you can type agent. So, now I have free access to this. So, um what what are we going to do? Um build a direct clone of Streamyard and make a marketing website for it. It's kind of funny because if you if you want to do design work, you should use Fable. But I'm just saying if you have some agentic work to do uh and you want to use a model that does anything you ask it to and doesn't like ask too many questions and just does it and there is no you know too much sense of moral, you probably want to go with this one. Yeah, I guess so. Yeah. Um I did try and generate a sexy dragon though uh cuz you can generate images and video from within Grock build. Mhm. It refused. So, sorry guys, I can't show you a sexy sexy dragon. There we go. It's going to build stage cast. This is So, we're using Streamyard right now. Um, and it's going pretty quick, I have to say. Well, they have all the infrastructure. This is quite good. And this is picking up on my like agents and my skill all my skate agent MD files and skills and things like that. So, it's able to because it that's the thing. Claude is the anomaly here that has its own clawude folder with its own claude MD and its own skills. It's so annoying because if everything just used agents.mmd and use the agents agents folder with all the skills in there, then you could just float in between all of these tools. But a um Antropic really are the apple of AI, you know. Yeah, this is this is pretty quick. Uh, by the way, are you using the free tier? Yeah, this is free. This is free. But it's $30 for seven. But you signed up for the $30 but with 7 days or No. However, I did have to sign in and I'm sure it's knocked me into that free trial tier because I have access to it and it's working because Yeah. Yeah. So, there is a 7-day free trial for me. Exactly. It doesn't allow me to do anything unless I subscribe to the seven uh now I need to tell an agent to remind me in 7 days to you know cancel to cancel it. Uh oh. You got a third subscription. Do you want to see our StreamYard clone built by Grock? 4.5 in uh 10 minutes and 9 seconds. You wanted a landing page. Here is your landing page built by Grock 4.5. It does look like Claude design. It's just this just shows how how prevalent this is on the internet, isn't it? This is this is, by the way, this is when I say like building 500 websites in an hour. This is the look that I'm going with. I'm not going with something crazy. This is this I know we hate it because we've been seeing it over and over and over again, but this is kind of like the winning layout that and this is the layout that 95% of the websites should have and they don't have it. Like just look at the old ugly websites. They they they can totally use this layout and it's readable. It's usable. It's not opening. It's not beautiful, but it's not distinctive. I mean, it's not ugly. It's when I say it's not beautiful, it's just not amazing. Anyway, so so yeah. Uh classic. Uh oh, we do have separate pages. Look, we do have separate pages. I didn't know that. So, yeah, we got separate pages. Whatever. More importantly, we open up the dashboard. Here we have a dashboard. where I have we've got some um destinations. We've got a brand kit. We've got recordings. We've got Oh, and and oh, this is what we haven't tested. Check the private messages. Don't reveal this URL. Just check the private message. Click on that. See if you pop up on the screen right now. Oh, it's not working. Well, because it is like on your local machine. How would that So what's happening? Uh it says this page isn't working. It's currently unable to handle this request. 5002. Meanwhile, you do that. Maybe I can show the design that Gro did for me. Yeah. So this is my website, my current website that we are like redesigning. Uh if you notice, we have like a hero section. We have this section with our past work. We are showing it off and we have like this section and I asked Grock literally here I said like create a clone redesign version of my my website homepage use webgl and creative coding kabarza.com. This is a very minimal um prompt. I wouldn't prompt like this normally but yeah that's what I did and here's the result. Let me reload. So, it's Oh, it's hard to see, but I don't know. It's not It's not amazing, but I guess yeah, I get the idea. So, it it didn't use my I I mean, I didn't specify anything. Okay. So, I'm not complaining about why didn't it do this and that, but it did use 3JS and Gap. Um, and yeah, I guess this is not the this design, but it is it is what AI does with a oneshot prompt. It's it's not too bad. Do you want to try that link again? Sure. Oh, yeah. I'm going to remove it is working. I don't see myself, but I can. Did you put your Did you type your name in? Yeah. Yeah. Oh. Um, you are you're in the green room. Oh, now it's asking for Wait, give me a second. Actually, I'm going to allow another camera just for fun. You're in the green room. Yeah, I can't see guests. Let me Oh, sick. Yeah. Okay. Well, maybe maybe we can't I I can't see. Whoops. You show the link. What link? That's fine. The the invite link. No, no, no. People can't get it to it from there. Oh, like scale link. Okay. Well, yeah. I mean, for 10 minutes, you know, there's a lot more that we would have to do to um Oh, I did a test whilst you were talking. I did a test and it records to your local machine and all the rest of it. I can download the local recordings. It's uh it's pretty good. This is part of a larger conversation on my show, Command AI, which we stream live every single week. We discuss the news and all things related to AI in the world of design and web. Catch us next week and join in the banter. Command AI.