China’s GLM 5.2 KILLED Claude?

So, China just released one of the most powerful models I've ever seen. It's right up there with Claude on my own tests. And what you can build with it is absolutely wild. I'm going to show you exactly how you can use it, how I'm building with it, and how to combine it with Claude code to basically automate and build anything. And it's outrageous what you can do. Let me show you some examples. So this is like a real full world game. I don't even understand the limits of this yet. And we fully automated it using GLM 512. We're just like, right, build me a game. Bear in mind, I'm non-technical. I'm not a coder. I'm not a developer. But we can build whatever we want, including full 3D real world games, as you can see, using this powerful system. Here's another example. So this is like a crazy game that you can build out with this stuff. You can build websites, apps, games, tools. We've even linked it to our memory galaxy inside the Agent OS. So this is one of the other cool things that we've done with it. Over here, you can see that this is all linked to my memories. And we have a galaxy of stars. They're all my memories and everything that's going on in my life. Here's an example. And so we can zoom in, zoom out of this galaxy. My AI agents can plug into it. GLM 512 can organize it and build it to get more context. And we can automate videos. We can automate SEO. We can automate music, et cetera, using this powerful system. We can even go from idea to implementation in like two single clicks using this powerful system. So how are we doing this? What happens here? What's going on, et cetera? So basically what we're doing here is we're using GLM 512. Now, something that's interesting to note here is that the same day, you know, Claude got banned pretty much. China gave them away. So Claude Fable 5 got banned and then GLM 5.2 dropped. And it's absolutely wild. Here's another example. So this is a full video made with GLM 512, right? This is fully automated. And bear in mind, like the video itself, we just type in one single prompt. It creates a B-roll. It creates the full video, as you can see. It automates it all, plugs it all together, puts my voiceover on. We can even add like an AI avatar to this too. And it's unbelievable what we can do here. Now, here's the announcement from GLM 512. So they said it's now available to all people on the coding plans, which means that you can actually plug it into something like Hermes Agent and then get teams of agents to build avatar videos like you can see. So if you've never used GLM 5.2 before and you're thinking, oh, it's probably not that good. It's probably trash. Well, it created this whole video in one single prompt, which is better than most people's videos out there. And it did it like, you know, just on autopilot. It's our age. Here's some other stuff that it's built, right? It actually built out a full operating system. So if you open this up, these are all different apps, including a music app, that we've put inside a completely different operating system, right? You can see, for example, it's created paint. It's created this notes pad where we can close it, then open it up and it saves everything. We've even got like the file and edit and view and everything like that. So full operating system here and then a calculator as well, right? So apps within apps, that's what we are doing. It's getting very meta like so. And we can also give this to Hermes Agent as well. So we can log into Hermes Agent using our coding plan, which you can't do with Claude. That's a huge advantage to using this new model from China. And bear in mind, this is a tiny little lab in Beijing that's actually created this. So it is Moonshot, Kimi Moonshot that created, sorry, ZAI created this from GLM 5.2. It's pretty powerful stuff. Here's some other things that were built. You might be wondering like, how does it perform? How does it work, etc.? So these are the announcements from GLM 5.2, Frontier Intelligence Openweight, right? Open source. So what's crazy about that is that this is fully open source, MIT license, which means people can use it commercially. And it's completely open. Now compare that to like the US government banning Fable 5. Totally different attitude, right? Totally different approach to it, which is kind of scary in itself. You'll see it came number one on design arena. So it's very good at UI. It came 10th on the agent arena board, so just below Claude. And it's performing very well on benchmarks. It's right up there with Claude Opus 4.8. Now here's one of the craziest things. You can actually get Claude to build directly with GLM 5.2. So you can see, for example, we've had a full conversation with Claude Desktop. And we got it to actually operate GLM 5.2 and then build it into our agent operating system. So if we go inside this section, we can chat with it directly here. We can also go inside Hermes and speak to it directly there. And we can also get GLM 5.2 to build things and then save it inside our workspace. So all this stuff that you can see right here was fully automated using a combination of Claude and GLM 5.2. And I think that's really the best way to use this is like you can get Claude to orchestrate all your agents. So if you get a new CLI and you want to test it out, just get Claude to run the test. Get Claude to build some cool stuff. Train Claude on how to test models. And then you can always test new things and build new cool stuff like you can see. I mean, here's a full open game that we built out. It's basically Minecraft, but we built it out with GLM 5.2. When I actually did the same test with other models, including Claude and also KimiK 2.7, it didn't look as good. And this is something else that's pretty interesting. Versus KimiK 2.7 versus Opus 4.8. Like the actual difference between these is massive, right? Like look at the quality of GLM. Sorry, KimiK 2.7, which is over here. Not very good. Look at Claude Opus 4.8. Not very good. But look at this one. Like full 3D game. Ready, playable, easy to use. And we could just go off and use it and build with it. So on some tests, not all tests, but on some tests, actually performs better and it's more fun to create with than Claude Opus 4.8, which shocks me in itself. But again, like best way to use this is to build with them all together. Now you might say, what about KimiK 2.7? KimiK 2.7 is pretty good. But if you look at all the tests that we've ran, like Opus 4.8 is pretty good. GLM 5.2 is pretty good. KimiK 2.7, it was never quite there. It was never quite as good. Here's another example. So this is Claude Opus 4.8. And this right here in the middle is using GLM 5.2, right? The quality of this game is much better than this game. And here's another example where GLM 5.2 actually won. So this is Opus 4.8. Pretty boring. We created like a liquid in a bowl test where we wanted like in one shot, thousands of particles sloshing in a round bowl till you tilt the mouse, right? Which we've done over here. And this works beautifully. Looks really nice. Pretty cool to use. This one with Opus 4.8 doesn't work at all. So you can see the difference here. Even like the website, I would say was better built with GLM 5.2. So it's right up there. I still wouldn't say it's as good as Opus 4.8. But on many tests, it's beating Opus 4.8. That's the point that we're getting at. And bear in mind like if it's open source, if it's open weights, if people can build on top of GLM 5.2, then it's probably a matter of time before they overtake Claude. Bear in mind Claude's models are getting taken down from the market. But no one's going to take down GLM 5.2's market, right? No one's going to take down a model from China. That just wouldn't happen. So it's pretty wild. And these are just all the cool things that were built with it. Now, the other thing I like as well is like whenever these models come out, we can just plug them and slot them into our agent operating system. If you haven't done that before, it's pretty good. Like you can build, for example, teams. And I only say this to inspire you, to show you what's possible. Because 99% of people are just building snake games and to-do lists and basic stuff like that. They don't even need frontier level technology for. But if you can have a full company of AI agents working together, and if they actually create stuff like you can see right here, then that's super useful. And you might say, well, oh, yeah, you know, AI doesn't create anything useful. This is AI slop or blah, blah, blah. Some people say that because they're not open-minded. But if you actually look at the traffic these websites get when we publish them that are built by stuff like GLM 5.2, like, I mean, look at this website, right? It went from zero clicks a day all the way up to 72, and it's growing with SEO. If we look at this website from, like, zero clicks a day all the way up to 222 clicks a day. So if you're thinking, oh, this is stuff that you can't really use for business, no, you're wrong. And you might say, well, okay, what about the AI avatar videos? Like, surely they can't, you can't do anything with them, blah, blah, blah. So if we have a look at this channel here, you can see that we've created a full channel around AI avatar videos, fully faceless, like this one, for example. And that's already grown to, like, 6,300 subscribers. The content is really good. It's better than I could create personally. And also, the other crazy thing about this is people actually really love it. Like, the comments on this are super positive. So, like, if you're thinking, oh, you know, video agents and SEO content and websites not useful, so wrong, so wrong. And I know, like, most people probably don't think like that. They're not closed-minded if you're into AI. But I do see sometimes in the comments, and I'm like, wow. Like, people don't realize the full power of this and how to use it, etc. So if you've not checked out GLM 5.2, it is absurd. What you can do with it, how you can deal with it. We can have, like, a full group of AI agents working together using this. We've got a pipeline where we go from idea to implementation. We have an agent Kanban where we can run local agents and just build them together. We have Claude, OpenClaw, Hermes, Gemini, Antigravity all built into one system. And then we have free Claude code operating on another Chinese free API, which is called N2. And you can just plug that into the system. You can build with your voice. You can create awesome stuff like you can see, right? We've even built, like, Fusion into the system. So we could have GLM 5.2 working with other AI agents. And when they work together, when you have, like, five models coming up with the same answer, and then you have a judge that fuses the ideas together, you need to actually get, like, Fable-level intelligence based on the benchmarks, which is pretty wild itself. Another thing that I've discovered recently that might be worth checking out for you is that we've created a loop engineering system. So basically, we give it a goal. We give it a max amount of rounds to go. We give it a free API. And then you can get, for example, like GLM 5.2 to judge the model over here. And so they self-iterate and self-improve based on the judge giving them feedback until the output is really good. So you could say, for example, I create an amazing blog about GLM 5.2. So the builder will start building based on the goal. And each round, they'll submit their work. And then the judge, which can be GLM 5.2, can judge that work, give them feedback, and be like, right, this is not good enough, or this is good to go. And once it's completed, it goes in our workspace, which is pretty amazing. So you've got the SEO section here. That video agent that I showed you before, the one, you know, where you've got the fully edited video of the AI avatar and the voiceover and the B-roll all plugged together with the research and the script automated too. That's inside this system here. You know, we're at the point now where you can basically automate anything. I have not hired for months as a team because we don't need people on the team anymore. We just need more AI agents and better systems. And that's what we're building with this Agent OS. So if you want to get the full system from me with GLM 5.2 and Claude plugged in, you can get that inside the AI Puffer boardroom. Link in the comments description or go to the AIPufferboardroom.com. And inside the classroom, you can get the Agent OS system that we've literally just updated today with a new zip file and a video tutorial and daily new tutorials like you can see right here. So if you want to learn and build with this stuff, you can get it all inside the AI Profit Boardroom. Inside the community, you can ask questions, get help and support whenever you want to. And one of the cool things about this is that we can have people sharing what they've built as well. So you can see, for example, Marlon shared their Agent OS system that they used from us. They tweaked it from us. Look how powerful that is, right? If you want something like that, you can get it inside here. If you want all of my best new trainings on this stuff, you can get it inside the classroom. And we have a beginner to expert automation course if you've never used AI before. Inside the calendar, you can jump on four weekly coaching calls, build this stuff live, etc. And then inside the map, you can actually meet people in your local area who are building with AI agents just like you using GLM 5.2 and Claude and whatever. So I hope to see you inside there. Thanks for watching. I'll see you on the next one. Cheers. Bye-bye.