Transcription
So there's a new method to make cursor 10 times better and almost no one is talking about it. But just 3 days ago, OpenAI made a major announcement that their own agentic AI CEX is now publicly available and works directly inside of Cursor, making your AI coding even more powerful. And as someone who has built applications for over 20 years, I can tell you that this is unconditionally the best AI coding setup available right now. And it costs as little as $40 a month.
I've been using this setup myself for the past few weeks on multiple of my apps. And my students who are complete beginners are also using it to build their first real apps in weeks instead of months. So in this video, I'm going to show you exactly what Codex is. How to use it and how to get this all set up in under 10 minutes so you can start building better and faster today.
So let's start by showing you what actually happened because most people completely missed this. So, Codeex has actually been available since May, but unlike Cloud Code, you couldn't use it with your Chat GPT subscription. So, most people never tried it because it would have been very expensive to do so. But then, OpenAI released GPT5 and they dropped the bombshell. Codex was now included with ChatGpt Plus, Pro, Business, Etu, and Enterprise. This was a huge deal for a lot of people, including myself. We all started testing and experimenting with it immediately. And not only was it good, it was great. GPD5 inside of OpenAI's on aentic coder felt fantastic. It was super autonomous. It could go away forever. And it did a way better job than Cloud Code and Sonet ever could.
But this still wasn't enough for OpenAI. They really wanted to crush the competition. And so they did. They gave cloud, they gave us the feature that cloud code users were asking for for months, a proper integration into cursor with their codeex IDE extension. But this still wasn't enough for OpenAI. So they trained a custom version of GPT5 called GPD5 codeex which comes in three versions and is clearly like one of the best coding models out there in the world and OpenAI made the most powerful one GP GPT5 CEX high available only in Cordex. So to this day this model is not available in cursor or anywhere else. But here's the thing. You actually can use GPT5 Codexi inside of Cursor if you know how.
And I've been using it a lot in the past few weeks to build my free AI coding prompt learning tool. Link in the description. I have updated Macropulse, which is a global financial dashboard that teaches you about the financial market without all the news that you know comes handinhand with it. And I'm currently building CourseFly, which is an AI native online course hosting platform. And you want to know a little secret? Even though I have over two decades of experience, I haven't written a line of code in 6 to9 months. And the setup I'm about to show you is the best one to do exactly that.
So, first things first, let's talk about pricing because be careful. This setup can cost you as little as $40 a month and as much as $400 or more. So, really quick, here is what the pricing looks like. So, Codex is part of the Chat GPT subscription, which means it comes in plus for $20, Pro for $200, and business for $25 if you pay annually or $30 if you pay for a year in advance. And then the pricing on cursor is also $20 for the pro plan. than $60 for the Pro Plus plan and the Ultra plan cost $200 a month. And here are the combinations that I would recommend depending on where you are right now.
If you're just starting out, get ChatGpt Plus Ed Cursor Pro for $40 total a month. This will get you started, you have full access to all the features of everything and you can really get a feel for it. But if you're building a few hours every day, two to three times per week, then get Cursor Pro Plus and stick to Chat GPT plus for a total of $80 a month. This setup will be more than enough for what I just described and you can always upgrade later down the line if you hit the limits because then you need to decide which workflow do you like more. Do you prefer the cursor workflow or do you prefer working in codex? Cursa generally has way more features, but it is more expensive. And Codeex has way fewer features, but it has way more raw coding power in terms of monthly usage allowance, if that makes sense. You can really just upgrade one of them, two of them, whatever you want. Just do it based on what your needs are. And you can also do both. Like I I'm on $200, you know, on both. And I know Rob, that's so much money. How can I afford this? Listen, if you want to build tools, a setup like this replaces a developer that would cost you thousands of dollars a month. So, the most important thing step here is not to think about like 40 or 80 or $100 or whatever, but that you start building first and then upgrade as time progresses, as you validate your ideas and hopefully start making some money.
And one question that I get a lot is why do I need both? Why do I need cursor and codex? Listen, cursor is an amazing tool. It's the world's best agentic environment for coding, even when you're starting out, hands down. But all of this is not cheap. And if you're not made out of money, you will want to put the heaviest thinking task into another AI so that you can use cursor when it actually matters. For example, to use models other than GPT5, which is what Codex uses, without burning through your very expensive cursor credits.
So, let me show you exactly what I mean because right now it is 11th of October as of recording and I am $380 95% into my cursor ultra limit for this month. This is not good. I have about $20 left for another 11 days. I did a lot of building and testing this month. So, this is just not a good position to be in. But there's hope because all I have to do now is go from the standard cursor chats into the codeex chat right here. And just like that, I can show my rate limits. I got lots of limits. I got weekly limits that reset in like 5 days from now, which I've used only 2% of. There's so much AI coding power waiting to be used. Plus, you get, like I said earlier, access to models that even cursor doesn't have access to. Specifically, GPT5 CEX low, GPT5 minimal, and most importantly, GPT5 CEX high. This thing is such a beast. It has built things. It has on occasions gone away for over an hour of just thinking and building and then it came back with something that actually worked.
All right. Right. So now that you have seen why this works, let me show you exactly how to set this up. So to set this up, just follow these three steps. Number one, you have cursor installed obviously. Now you need to go to the marketplace, the extension marketplace up here, and you just click on it. And then you need to be a bit more careful because there are a lot of people that are trying to be found with their extensions when you look for codeex, which is not okay, but it's life. So you see if you look for codex you can't even find it because everyone is trying to be on top like search engine optimization. So what you have to look for is like open AI codeex and then maybe you'll find it. You'll see there's a big difference in what it looks like when you actually find it. So now okay so here you see now now it's number one but it's not always number one. So, just make sure that you download the one with the verified check mark right here showing OpenAI, right click on it and then install it. This will only take a couple of seconds and then you'll have it installed. It will usually launch automatically like it does right now. Sometimes, like right now, it bugs out a little. It's not the first time I've seen this. All you have to do here is just cancel it. Like I mean, not cancel it, like close it and just restart cursor. It happens I don't know like one out of five times for some reason. But now it will launch and now two things will have to happen.
Number one you'll have to sign in with your chat GPT account. Do not use an API key. You will pay for your usage if you use an API key. So sign in with chat GPT. So we'll do that. It will ask you do you want to allow cursor to open this website? Yes. So we'll just open it. And now you'll have to log in which I'm going to do real quick. So once you're logged in you'll see the screen right here. here it will show your email address and whatever and you just say continue and once you click it cursor will open again and you will see that now you are indeed logged in so now there's a little bit of a tutorial saying codecs in your IDE cools in the cloud cool that's for another video to-do list cool for another video and yeah cool let's go so now you can use it this is literally it that's it man like from here you can like go all in there's just one more thing that we need to do now and that is unless you are a psychopath that likes to have their agent coder on the left, we want to move it to the right, right? But if you go to this icon up here, which would be, you know, what what people would usually expect, and you just try to like move it, it doesn't work. That's weird, right? So, what we have to do go to this codeex label right here instead. And then what when you drag this, it actually works. So, now you can move it wherever you want. for example, right here. And here you have to ask yourself, do you want to have it as the primary coder inside of cursor or as the primary coder, right? Uh the secondary coder. So I like to have the cursor chat first and then the codeex chat. And now that's really it.
So now I can just go here. So the new cursor chat, you already know this of course. And here you see you have all types of models including all the GPT5 models, but not GPT5 codeex high or low. And then here you have all of them, right? Like GBD5 cordex low, GBD5 cordex medium high, minimal, all of the GPT5 models. So let's just go with GPD5 minimal because it's the fastest model. It's not the best by far, but it's a fun model to test things with. Uh we just say something like, explain this codebase to me by just reading the documentation and then summarizing in two sentences what this app actually does. And now you see it was just here and okay I I started I started talking too quickly explain in this codebase to me by okay that's enough AI are smart so I just send it just like you would with cursor and it always takes like a few seconds to send the message for some reason my internet is also a bit slower today for some reason and now it would start thinking just like cursor would it has access to the same files that cursor does you can see it working you can see the context window down here even like 5% used right now. It is pretty cool and again you can you can see your actual rate limits which on plus will give you like a few hours of coding every day right nothing crazy but now you see cursfly is a modern nextjs learning platform that lets admins build video courses manage lessons and invite students while tracking real-time progress etc etc that's it so now what I could do here is just go to GPD5 codex high and say and talk about a new feature that I would like to develop then it will come up with a huge plan really outsourcing the thinking if you if you like to think about it and then I could say okay cool approved just write this plan into a file like a documentation file or PRD or whatever you want to call it and then you could switch back to the cursor chat and say something like okay implement the plan that is outlined in this particular file for example this will save you so much money and will lead to much better results but before you run off and try this there are a few common mistakes that you need to avoid.
Number one, don't use codeex for everything because it is very very slow. It is very smart, but it is so slow that you might lose your motivation to build. I like I know a lot of people don't like GP5 because of that. So if you just want to for example change the color of a button or text inside your app or ask a question about your app then going to cursor and just asking it with a model like 4.5 sonnet or maybe even the new stealth model cheetah will lead to much faster results. Mistake number two is don't overuse cursor either specifically with the models that are very expensive. So if you look at this is a real month of mine. If you look at the breakdown over how much each model cost, you can see that Claude 4.5 Sonnet Thinking took over 50% of my monthly cursor subscription. And it isn't it isn't even like for that many tokens. If you compare it, it's about three times as many tokens as I had with GPT5 fast. And GPT5 fast would have cost me about $120 in comparison versus $232. So be careful when using sonnet especially with the thinking models sonnet or oppus avoid oppus altogether actually. And then number three don't use not in cursor and not in codecs the minimal or low options unless you know why you have to understand that GPT5 is a thinking and reasoning model. Right? It shines when you give it the time and space to just think about things really analyze it. So by using something like minimal which is much faster but just doesn't think that much or even the low models unless you have a plan will just lead to bad results. So I highly recommend that you make use of the high reasoning capabilities of GPT5 and then if you feel like it is too slow for the way that you like to work then make a plan and give it to Sonnet or maybe even Cheetah.
So, now you know how to use cursor and codecs together. But here's the thing. Your entire setup is useless without MCP servers that make your AI agents 10 times faster. Most people don't even know these exist. So, watch this video next where I'll show you the eight best ones.