New Grok Build is ABSURD!

**Julian Goldie** (0:00)
For the first time in my life, Grok Build actually seems good, and I've never seen this before. Now, I always wrote off Grok as one of those models that was pretty subpar, and tried to do everything with Twitter and everything else. But actually so far, it's pretty good. I mean, we built this, and this outperformed many of the other stuff that were built using the same prompts on other models, especially the ones that recently came out. And it's mind-blowing what you can build with this. Plus, if you already have a Twitter subscription, this is great, because you can generate images, you can use a CLI, you can code out whatever you want. I haven't hit any token limits. We can build all sorts of crazy stuff with this, and it just comes directly with my Twitter subscription. So if you're in the same situation, it's pretty good news, right? So I'm going to guide you through what I built with it, how I'm building with it, how I see this stuff going.
Some of the cool ways you can use this as well, especially agentically, and then what I think overall is the future for stuff like this. So let's get into it and stop playing games here. So if you're not familiar with Grok Build, basically, if you already got a Twitter subscription, you can just use this and plug this into your system, so you don't need to pay for subscription or APIs or anything like that. And with Grok Build, you can use it inside the CLI, but you can also plug it into your AI agent. So if you go to Hermes, for example, over here, we can use this studio and I can switch between Grok and Minimax, right? Minimax M3, another Chinese model, pretty good. Grok Build is inside here, and we can automate images, videos, whatever we want using this section as well. We can even create voice notes, which is pretty cool. And then not only that, we can actually talk to our agents on a live call, which is pretty amazing as well in itself. So when you're using this, there's actually multiple different ways that you can unlock a powerful CLI without having to do much extra work or pay much extra for it. Now, is it up there with Grok? I'm going to show you some examples and we'll compare it side by side, but it is good. Like it's, I mean, we built this in like a couple of hours using Grok and it was super simple and easy. Plus, I mean, look at this. This is fully open world. Like this is a full game that we've built and I don't code. So how amazing is it that you can build all of this stuff and you don't need to be a coder. You don't really need to do anything special here. And this game seems to go on forever. So you can see these two games we actually built directly with Grok Build. And the great thing about this is it scores 70.8% on SW Benches, which is not pretty not bad at all. 100 tokens per second when we actually measured it. You can actually have eight subagents working in parallel when you're using Grok Build. And it's a terminal coding agent as well. So it's pretty good. Now Elon Musk was actually asking for feedback on this. And I know why, because it has got dramatically better. Honestly, Grok itself was trash before if you ever used it. Now it's a whole lot better, my friends. And also what's interesting about this as well is like you can use X search if you plug in Grok into, for example, OpenGlor into Hermes, you can do Twitter search here. You can see some of the searches we've done previously. It can search web, but the great thing is it can search Twitter too, which most LLM struggle with because they don't have access to that data. This is like the only LLM that actually has access to live data on the most popular news website in the world, aka Grok, sorry, Twitter, which is pretty amazing. You can also talk to Hermes and OpenGlor and it's fun to use. It is a lot of fun. He's got a whole lot better recently. Now I want to show you an example side by side of like what we got and how it performs.
But basically you can have like eight subagents with Grok Build working at one time. So you can basically give it an order, you'll plan it. You'll have like subagents building in parallel up to eight of them. Then actually self checks his work. This is built into the CLI. And if it's good, it will ship it. If not, it's going to do it again, but sharper and just keep going around as a positive feedback loop. So Grok Build, it doesn't just like start typing and building. It always plans the job first. It splits it across subagents to work at the same time, checks its own work, then ships, then it loops. That's why one sentence, which you've seen before comes back as like something super nice, which is pretty awesome in itself. Now, the way that I would use this, because I don't think it's as good as Claude 4.8, is that I'm going to plug it into the agent operating system, and then when I need to build out cool stuff and I need a lot of tokens, and we need to blast something at our agents, well, we can do that inside the agent operating system. That's what I'd recommend for you. It's like you probably wouldn't have this to replace Claude, but you would have it to build stuff, especially if you run out of tokens on the Claude plan. That's a good way to use it.