The BEST AI Model For Developing in 2026
Let's learn how to use Cursor AI📅 2025-11-04
Transcript
[00:00] Which of these models in Cursor AI is best for vibe coders? And I'm going to answer that today. Back on today's video, I'm going to go over the best model to choose if you're a little bit more experienced or you just kind of want a model that's going to give you code correct every single time with limited errors. As if you're more experienced with software development, I'm going to go over some methods when it comes to granularizing the way you approach different logic specific to your application. But for now, let me just answer that question right off the bat of which is the best model if you're a vibe coder. And you can already see it selected. is going to be GBT5 codeex
[00:31] high. This model is probably one of the most powerful models I've seen in coding in a while. And this model takes very long to do task, but effectively the code that comes out of that task is very good. Now, let me show you how to enable this. First off, go up to your settings. You're going to want to come to your models tab and then if you don't see it, simply search it here. But as you can see, it's enabled. Boolean on. Good to go. So, I'm not going to make you wait 5 minutes to give you the answer. It's GBT5 Codex High. Now, I'll explain the nuance. Now, if you're a little bit more experienced in developing, which model you should choose for what. If you're
[01:01] completely inexperienced or have little to no knowledge on coding, opt for this. Now, if you're a little bit more experienced and you kind of understand the implications of what you're requesting in a code job, this is where choosing different models comes into play. So, what do I mean by this? If I'm going to do something extremely simple, for example, I look at my header here and I'm like, you know what, I need to change the color of scroll stopping or you know what, I need to change the copy here so it doesn't say design stop scroll stopping thumbnails in seconds. It says thumbnails are amazing. I would just opt for auto to be honest with you.
[01:33] We just need it done fast. Give me the code. Easy. This is lowlevel logic. When do we opt for something higher like GBT 5 codeex? This is when we do logic that is going to take the agent 30 to 40 minutes to even produce the value. So an example of that for my platform which I knew when putting in the task for it is going to require a higher level of reasoning was its ability to handle this workspaces feature I'm currently developing. Idea being that each individual that signs up for this platform is going to have its own associated workspace. But on top of that
[02:03] with the associated workspace there are going to be relevant seats remaining. So, if I come down here to pricing, what you'll notice is that we have our nice little tiers. It's amazing. But each tier has a relevant amount of seats, such as each seat is $24 a month if you on the monthly plan. Therefore, to add another increment, another seat be $48 a month. I knew this logic is going to be way more complex than just simply changing the color or the copy of a landing page. Therefore, when I approach this kind of logic, you want to approach it in ask mode, GBT5 codex high. You
[02:36] want to talk back and forth to really find the road map that is most optimized for your specific software. Ask mode is extremely powerful when using a highle reasoning model like GBT5 codex high due to the fact that you are able to ask very specific questions that is specific to your underlying software's front end backend and long-term architecture so that when you connect this to a real backend or you have a software that functionally has users you're not shooting yourself in the foot when it comes to poorly designed architecture that all sounds super ambiguous I know
[03:09] so you got to watch another series on this channel I am making hour-long episodes here y'all. It's in the description down below. It's about Google AI vibe coding. I show you how to vibe coding real software from a real engineer's perspective step by step. So, if you're more skilled, you want to granularize the way you use these agents. So, typically if I want something done that is more mid-level of complexity, maybe not high level complexity. I'm opting between set 4.5 and GBT codeex. I use these interchangeably. Personally, if I'm working on a job with GBT5 codeex, I
[03:39] notice it's taking longer than it should or I'm noticing it's hitting Roblox, it shouldn't, then I will put in the exact same prompt or alternatively say continue the job from this model and give it to sign up 4.5 to give a different perspective. What's occurring here, which is really cool, is that if you were running a real software company with a real software team with real humans, I know it's kind of crazy. Essentially, what we're asking here is a different perspective, which is very important in coding. One thing you need to understand about developing software is that when you find the end value of a product. So for example, mine is editing thumbnails. The way we approach editing
[04:11] thumbnails through AI. There is multiple ways I could have gone about the pipeline to upload the thumbnail data to analyze the thumbnail data to then edit the thumbnail data. When I say thumbnail data, let me make this like make a little bit more sense. Eg, you know, log in, continue, you know, I drag in just a starter thumbnail here. put in a problem like making me hold a burger in both hands. It's amazing. This logic editing the thumbnail, cancer generation, uh this being sent to a Gemini API endpoint. All this there is multiple ways to approach it. Nice burger. What
[04:43] I'm telling you is that you want to get perspective from all these different AI models on the ways to approach it because it's like asking your team for a different perspective where you have a nice little whiteboard and you're drawing out different types of architecture. A specific type of architecture that is very important for you to understand is data paths. I'm not going to bore y'all, but what I mean by data pass is that when a real user signs up for a software, we will use their login information. We'll use their email and then typically we will create a thing called a UID. What a UID is is a
[05:14] random string. What a random string is is just a random list of characters just numbers and the alphabet. This UID will associate to that underlying user which will then create what we call a data path. These data paths are going to have relevant information specific to that user. So when we reflect their information on the front end, we know who we're reflecting it to. We know that this creator at thumbio.ai, this individual has created this project because we stored it on the correct data path. I bring this up and I dive into it way more into that series is that this
[05:45] is extremely important for you to understand when diving in between different models and when creating more complex logic. So TLDDR of what to choose for what I would choose GBT5 Codex high for planning and executing extremely complex task. I would choose auto for very very very simple task that you know you could do yourself basically in 5 minutes but why do it and just have AI do it? I would choose 4.5 and codeex for midle of complexity of task. What's an example of mid-level Corbin? You know it's really dependent. Uh you know one one example here could be you know look at this sidebar here. How do we make the
[06:17] UI and the sidebar better? Do some planning and then let it execute. uh it's really specific on your platform. But what I can tell you as the end idea of this video is that if you're extremely inexperienced, brand new to coding, you don't even want to look at the files, you just want to be in agent mode, not editor mode, use GBT5 codeexi. Is it overkill? Yes, it very much is overkill. But will it give you correct code every single time? I would say 90% of the time. that extra 10% where you may lose the code or the code might not be good but you keep iterating but you
[06:48] keep building you'll realize in like 3 weeks oh that actually wasn't good code and you'll adjust. So as you already know with these style videos I'll see you in the next these AI models are getting so good that I guarantee you by the end of 2026 anyone's going to be able to code anything in the entire world because of the fact that these models are going to be so good that they're going to be able to self-correct and find all that little nuance it doesn't even matter anymore type of video. Nice burger.