GLM 5.2 Is Actually a Game Changer

So, I've been playing with GLM 5.2 and Kimmy K2.7 code, and I have to say they've both impressed me so much. And I think I think we've reached a point where Chinese open- source models that are cheap are actually able to do real work. Command AI, Fable 5 level, not Fable 5 level. And this is what's this is what's crucial about that. This is what's crucial about this. This is the sorts of prices we're looking at. $12 a month. This is you're not really going to use this. This is for anyone doing any real work. You're not going to really want to you you'll reach your daily limit like or or at least your hour like 4 hourly, 5 hourly limit in in an hour. Um realistically, we're talking about the pro or the max. And if we're going to go for it monthly, it's $64. Quarterly 57 and then yearly is is $50 a month sort of thing. So really really quite a cheap alternative to something like Claude. I got it to redesign an app for me cuz apparently it's it when you say Fable, it's apparently doing better than Fable in design benchmarks. I want to see that. Let me show you that design benchmarks. We're not all about benchmarks on this channel. Oh, it's my own post again. I didn't plan for this, but my own post is coming up in Google when I'm trying to search this bloody benchmarks. So, it's coming up top. Can you see this? Um, again, I'm single paging right now, so I can't see. It is coming above Fable on the uh design arena. So, I put this to the test. I used it from open code. So, I didn't run this locally. It's like a 1.5 trillion parameter model. I think someone someone's quantized it down to three bits and it's actually still really good, which gets it to around 250 gig. I can't run it on my Mac, but 250 gig can run on a Mac Studio, you know. Um, and the performance is is quite good. I can't show you the leads page, but it's not it's not really all that bad. You know, it did have it started with a really clawed design first. So, there was a few parameters it sort of worked within, but I but I gave it a I gave it a um prompt that I have given Gemini in the past, which is here's an app that I've got. Don't change anything except just redesign it. Don't change any of the layouts, though. And I I have to say I was more impressed with Gemini than I was with this. It It's arguable cuz this is such a boring app. It's it's an outreach tool, you know, it's a basic dashboard, but um all that to be said, I added some features. I uh you know I changed some code. It got it done really quite fast. It wasn't like it didn't offend me whatsoever. It was it was it did what I asked it to do. And like for the first time I was using this a Chinese model and it just seemed to get done. And so where I think this is where I think this is a real turning point is Opus Fable, you know, for real research and creating clear implementation plans for a lesser model. And then you've got the Chinese models working away getting getting the real work done that I want to put this workflow to the test and see how useful it is. But it I was like it was just really satisfying knowing that it was barely costing me anything. And I was like, "Oh, I think we've reached Opus 4.5 territory with Chinese open source models now." And I think they're really usable. But as with all of this stuff, I got, you know, people slating me in the comments, these are rubbish prompts and this is not good or whatever. You all have your own use cases. You all have your own styles. I strongly suggest putting these to the to the test. Um, I've bought myself a year's subscription cuz I've got loads of referrals. Actually, I'll put I'll put my referral link in the in the description. Can you send some of those to me? Tokens. What? Credits. Tokens. Yeah. Like gifting each other tokens. I'm going to give you 10% off. I'm going to give everyone 10% off uh off GLM. You know what I would suggest is buy the light plan for 16 bucks. Give it give it a light bit of usage over the course of a week. Don't push it. Don't go too hard with it. Just give it a whatever. It's cheap enough to try out. All in all, really happy with both GLM and Kimmy. Both of those both of those new models are doing really really well and uh I I had a lot of fun using them. This is part of a larger conversation on my show, Command AI, which we stream live every single week. We discuss the news and all things related to AI in the world of design and web. Catch us next week and join in the banter. Command AI.