**Julian Goldie** (0:00)
Glm 5.2 destroys Claude code. So today we're going to be testing them out side by side, seeing how they perform. This is actually built with Glm 5.2, as you can see right here. So you can build out like some pretty amazing stuff. Here's another example of what we created. We even built out a full 3D game, as you can see right here. And there's all sorts of interesting things you can do with this that we're going to run through today. And I'll show you exactly how it works, plus how it performs versus Opus 8 And the interesting thing about this is, if we have a look at the benchmarks of Glm 5.2, which just got released, so we'll run through everything today. We'll compare them side by side. And I'll show you what we've built and how they compare versus Kimi K27 and 4.8. So this is the new announcement.
And bear in mind as well that this is open source, so it's available on Hugging Face. That means people can run it for free. It's MIT licensed, open source project, and you can also run it on the coding plan. The other difference between Claude and Glm 5.2 is that Glm 5.2 you can actually plug into Hermes or your other AI agencies in the coding plan, which is pretty amazing in itself. You can't do that with Claude. You can't use that inside your agents without paying for the API, which obviously costs a lot of resources. And so you can see here it's hosted on Hugging Face, it's available, and also on Design Arena, this is pretty shocking. It is scored number one on Design Arena. So it is outperforming Claude Fabel 5 on Design Arena, which is wild when you think about it. Claude Fabel 5 was so powerful, people took it down, right? It literally just got taken down this week. Glm 5.2 doesn't matter, mate. We've plugged it in. This is a Chinese model, by the way, it comes from Zed. Claude Opus 4.6 is number three on this list, and you can see how it goes down. So, you know, GPT 5.5 is not even in the race here. And everything that we've built out of this does look pretty nice. I'll come on to that in a second and we'll show you some tests. Also already on Olamma, which is great as well. So you can use it on Olamma and we can see how it holds up on the benchmarks right here. We can cover more on the benchmarks. I just want to show you what we've actually built with it, because it's kind of fun to look at this stuff straight away. So we have Kimi K27, we have Glm 5.2 and we have Opus 4.8. Which one built the best first person Raycaster engine, which is kind of like a textured maze essentially. So let's run through these and we'll see how they perform. So first of all, this was Kimi K25, it's actually super smooth like when you go around here. Look how smooth that is when you're using it. It's pretty nice, right? So this is another alternative that's open source and just dropped from China. Then let's have a look at this one. So this is the option from Glm 5.2.
It's not bad, but at the same time, it's not great. So if you have a look, you can see that it kind of feels a little bit buggy when you move around and it's very hard to control that as well, right? But it's exciting, I mean, it is cool. Now, if we have a look at, this is Opus 4.8, and I would say this probably created the best one. It's still a little bit hard to control as we're running through, but I would say that is probably the best option. So far, I would say Opus 4.8 wins on the first version, and Kimi K27 nailed it as well. Now, what we actually did is we leveled up the Raycaster, and we added some interesting stuff in here. So if we open this up, as you can see, this is the output from Kimi K27. Kind of reminds me of like a Mega Drive game back in the day, and you see an example right there. Pretty hard to control, but it looks cool. Now, this is the one from Glm 5.2, actually pretty good, right? Actually pretty good. I would say it nailed it, to be honest.
So if we go through here, we can also see a map in the top right as well, which is super useful. And then if we turn around, we will see things popping up as well as we go along. So pretty cool game right there.
10 more minutes of transcript below
Try it now — copy, paste, done:
curl -H "x-api-key: pt_demo" \
https://spoken.md/transcripts/1000651996090
Works with Claude, ChatGPT, Cursor, and any agent that makes HTTP calls.
From $0.10 per transcript. No subscription. Credits never expire.
Using your own key:
curl -H "x-api-key: YOUR_KEY" \
https://spoken.md/transcripts/1000773226677