This is Why I'm Canceling Claude Code

I mentioned I cancelled my subscription to claw code because I want to commit to GLM 4.6. Now this is an opensource coding model which you know it beats out um it beats out um 4.5 which is where am I looking at? This one here in a lot of them. It's it's not on the Swedbench verified there. So, and the terminal bench, but the pricing now is nothing compared to Claude and I want to commit to doing it and and I have seen the reviews and I have seen people use it. It doesn't quite produce like the best or whatever. But the problem with a lot of these influencers and things like that is they're looking for the one shot because they know that your attention is going to be lost if you if you say, "Look, you've got to work at it." Because surprise, surprise, guys, you have to work at success. It doesn't just happen, right? You've got to work at a code base. You've got to give it follow-up prompts. You've got to break it down into smaller chunks. You can't just oneshot an entire website with just one prompt. Do you know what I mean? Well, actually, let's go through the features. It's got a longer context window jumping up to 200k, which context size, I think, is a good parameter to kind of, you know, make sure you're being efficient with your with your prompting and your tasks that you're giving it. good coding performance. Uh, and you could actually trick claw code into using this by the way because they use the same API as anthropic, but I'm using it in kilo code. Um, reasoning more capable agents and uh, refined write it refined better aligns with human premises. So, I think that's Oh, no, that's that's actually so it can write and do your things. So, you might consider this for your script writing of your um of your marketing tasks. Yeah. 50 cents in $1.75 out down from three to 15 you know so you might find that this model it works for you this is quarterly so look at monthly $3 a month you know down from a hund I don't think this is more comparable to the $20 a month but um this is this is more compared to the $100 a month so it's 15 bucks per month but if you go you know quarterly or even yearly this works out about $30 per month to get the absolute max from this model, but I'm going to try in the API pricing inside a kilo code first. See how I get on with it and I might just get on fine with it. Are you changing models a lot? I used to swap between Opus and Sonic cursor to be GP 5lexity. I just kind of use the default to be honest. So I yeah, I've sort of found my rhythm a little bit. I just basically use CH GPT for almost everything. And when I'm inside of like cursor, which I I I spend quite a good amount of time in cursor, I use auto auto mode. But when I get stuck, then I try to see how different models compare. Well, you got 20 You got 20 bucks a month. You might as well use it. I have two cursor cursor uh subscriptions. One for myself and one for my developer. You might just give him an open router API key and put it through open router. might end up being cheaper than just paying a flat subscription. Well, one of the subscriptions I'm just I I used to be paying actually no, I paid it yearly because somebody asked me to build them something and I used the money to pay for the subscription for the whole year and then uh and then that deal came up and we bought basically every AI tool for the price of one. So yeah. Yeah, that's why I have No. Yeah. Yeah, you might consider because in cursor you can give it an API key to open router. I didn't know that. And then you can select any model and that just might end up being cheap. I don't know how often they're developing. I don't know how heavy they are. But maybe you cancel their subscription for one month, trial open router and say just stick this API key in. Then you can manage their usage. Then you could put a cap of $20, right? Um and see if they come close to it. they're like coming in at $5 then it's like cool use open router. I didn't know but I could do it that way but that is probably smart to do it for next year. Anyway, I love all you know I love trying new tools. I love I love um seeing what it can do for me and stuff like that. As I say 4.5 I had a rough first day uh 4.5 just nothing was getting done. There's no surprise that today wasn't a great day for me either. And I did try a few other models. So again, it's probably just tiredness, whatever it is. So I'm not I'm not going to outwardly say 4.5 is is bad and people have had success with it. Um I just implore people to make up their own mind on some of this stuff as opposed to just listening to dumbass influencers like me. Um should we move on? Yeah, we're going to warn you guys now. This is not a political discussion. This is not a political podcast. 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.