BOYS

BOYS Qwen 3.8 3.8 Max just dropped and it is amazing. And I know that I said this a lot, but this is actually that important model that I want you to listen for two seconds about. First things, it's 2.4 trillion parameters. To put that in context, K3 is around 3 trillion parameters, which does normal people, doesn't really matter, but the bigger the better for this. Really, that's all you really need to know. It also did on autonomous coding session for 10 plus days and they shared the GitHub for that, which is honestly amazing. Aside from that, this is going to be an open weight model. It'll be released next week. K3 is also an open weight model with around 3 trillion parameters. And they release their open weights the next week as well. For those who don't know, open weight is not open source. Open weight is a little less open source, but making it open weight does reveal a lot of sauce that goes into these models. All American based large models are not open weight or open source. Their benchmarks. This shouldn't yourself be as smart as the leading three or four other models that you're used to. 5.6, 5.6. And nobody else loves talking about it, but I love talking about Unsloth makes weight models that can be run locally. Smaller, they're dumber, they're slower, but you can still run them. So what they're saying is they'll be able to run this version of 3.8, not the full one, on 70 gigabytes of RAM. That's crazy. You could run this kind of effectively on a 24 gigabyte MacBook Air. For benchmark and I love the Arena and I actually honestly trust these. So you can see it's number four. It's well behind Opus 5 Max, it's behind Kimi K3 Max, it's also behind Opus 5 on high, and Fable 5 on high and 5.6. So they're smart, maximum thinking, but it is definitely smarter than Fable 5 on maximum thinking. Right. So right now the only way to use this model is accessing through their API, which I recommend you do, it's also pretty cheap. But once again, what this means for the wider AI market is very interesting. Is basically this is another Chinese model that will be open weight that is smarter than all the other cloud American open weight models that are not open weight. Had which Chinese and tested higher than every other American or Chinese model out. Have 3.8, the same exact stuff, but still not as smart as K3, still Chinese, weight. So a couple weeks ago we had Kimi K3 come out, which again, Chinese, open weight and tested higher than every other American or Chinese model out. And now we have 3.8, the same exact stuff, as K3, Chinese, weight.