China's Qwen 3.8 Max AI Just Dropped

**Julian Goldie** (0:00)
China just released Qwen 3.8 Max, and this is frontier level AI that's going open-weight. So this is 2.4 trillion parameters just dropped, and in some of the tests, actually coded autonomously for 10 days. So 10 days of self-evolving development. By the way, this just dropped like less than an hour ago. So this literally just dropped, and it is performing really well on the benchmark. So I'll come on to what I've built with it so far, because I've been testing it for weeks in preview mode. So it autonomously coded for 10 days, real work, long horizon mastery. So it can basically do all sorts of cool stuff. You can also see what it's created. So you can actually check it out on GitHub, and this is a fully autonomous GitHub repo the coding agents have been building with. Also, it's got multimodal intelligence, which means it's multimodal basically, and it can use different formats. And you can get free access at chat.qwen.ai. So you can get free access straight away to this, which is pretty awesome. And I'll show you some of the cool stuff we'll build in a second. You also get the API access on Qwen Cloud. I'll show you another way to get access in a second. And if you're wondering how it performs, very well indeed, my friends. It's also got more information on like coding tests and all sorts of cool stuff they did, but it is performing very well here. Now, I'm not just going to rely on the benchmarks. I'll show you what we've actually built side by side versus other tools in a second. But if we have a look over here, we've got Fable 5 on the benchmarks. It's not like compared against, for example, Claude Opus 4.8 or some old boring model. This is frontier level. And you can see when we look at the benchmarks over here, the SWE Pro, like it's right up there. It's right on par. In fact, it's actually beating Fable 5 on many of the benchmarks. But the craziest thing about this is, number one, it's Chinese, and number two, this is open source. So it's not just like, for example, you're getting frontier level AI from China. No, no, no, this is like going open weights. In fact, they're actually going to be releasing a local model for this next week, which is going to be pretty wild in itself. Now, if you look at all of the benchmarks here, it's performing very well. If you want to indicate how to use it, so what we've actually done here is we've already plugged it into Claude Code inside our agentic operating system. You can get that link in the comments description or go to the aiproperborn.com. You can basically get Qwen 3.8 alongside your favorite AI tools inside Claude Code. So Claude Code is the harness here. And then, for example, we've got Qwen 3.8, we can plug in Kimi K3, and then we can use the best agent harness in the world with one of the best frontier models in the world as well.
And let's have a look at Goldie Bench. This is where I actually test all of these models. And when we were looking at the Qwen 3.8 Max version, it performed very, very well. So let me show you a few things that we actually built with this. And by the way, if you want to test these creations out yourself, you can check them out there as well. But its build is super nice.
You can create 3D games, you can create websites, apps, tools, landing pages, whatever you want. The other thing that I would say is it's got a nice vibe when you create with it. And also, when I actually did the same test with, for example, Fable 5 and that sort of thing, the problem with Fable 5 was quite often the tests were quite linear in terms of what it built. But this one, it feels like a bit more open. And the vibe and the detail of what it creates looks super cool as well. Here's another 3D game as you can see here. This one was a bit more basic, so I wouldn't say it performed as well as Fable 5 on that directly. This was pretty cool as well. This is kind of like a 2D game. We've got this racing game as well. But yeah, it can create super nice stuff.
Also, you might say like, why would you just test it on games? But the thing is, this gives you an idea of like the physics, the coding, the logic, and also how well it designs UI. So I don't usually create games, but this is really just to test it out and see the full limits of the model itself. Here's another example. So it created like a kind of Skyrim open world game. As you can see right here.