Did GPT-5 Codex just kill Claude Code?

Open AAI just released a new model that changes everything. It is called GPT5 codeex and it is a direct competitor with clawed code and claude sonnet and honestly it might just be better. This is a brand new revolutionary AI model purposebuilt for agentic AI coding and it brings a ton of advantages we're about to go over. I have been testing it all day building some pretty incredible apps and honestly it is amazing and it might just be better than clawed code. But is it worth dropping your Claude Code subscription for? We're about to find out. In this video, I'll cover what this new model is, how it works, how you can get it set up, give you some master tips for it, then put it through its test to see if it's better than Claude Code. We'll build some pretty incredible apps. By the end of this video, you'll know what Codeex is, be an absolute master, and know how to build amazing apps. Let's get into it. So, OpenAI just

released this article today and I'll put it down below that goes over what GPT5 codeex is. It is their brand new agentic AI coding model which is really exciting because up to this moment GPT5 in my opinion just did not come close to Claude Opus or Claude Sonnet for coding at all. It just for me it was unreliable, buggy. A lot of other people swore by it and said it was better. At no point did I think GPT5 was better than Claude. But GPT5 codecs now this is a different story. Here is a really important chart I want to show you that I think demonstrates the power of codecs. You know this is one of the big advantages that we'll go over. The other big advantage will be how you use it inside codec and the different ways you can kick off AI task for coding. But this is important. Look at this chart. What this chart is basically showing you is how many tokens the model uses for different length tasks. And what you can see here is for tasks that are really quick and short, it is much faster than GPT5. And then for tasks that are much longer and bigger in scale, it actually uses many more tokens. And why this is actually a really big positive is basically what this is saying is the model's a lot smarter at determining how much time to spend on tasks. For really simple tasks, it goes a lot faster and gets it done. So if you want to make a fix in your code, if you want to change a small little feature, if you want to fix a bug, it does it faster than ever. But if you're building a really big feature or a fullscale app, it spends a lot more time thinking and does a lot better of a job. In this video, I'll actually go through building out a fullscale app, a codeex, but this is really cool because it is much better at determining how much time to spend on a task. it can determine is it a quick task that we should do really quick or is it a long task that we should put a ton of compute behind and that's really powerful. The other really powerful advantage to this new AI model is the fact that it can hand off task. And basically what that means is there's a lot of different ways you can access this model, right? You can access it through an IDE extension inside Visual Studio. You can access it through codec cli in your terminal. And you can access it on the web inside a chat GPT. And now what you'll be able to do is if you start a task say inside a cursor inside the codeex IDE extension, you can then take that task as it's working and pass it off to codecs in the mobile app. So say you're building out a brand new feature using this new model inside a cursor inside Visual Studio Code and you're about to go and you have other things you can do, you can pass it off to chat GPT on your phone. If you go on your phone inside Chad GBT, there's actually a little button for codeex and this is where you can control your AI coding model. So you can start tasks on the go on codecs on your app. Then you can come home and then go into codeex on your computer inside cursor and continue that task on your desktop. What makes this really cool with this ability to go on mobile and kick tasks off wherever you are is it basically becomes like your own AI coding employee. So imagine this scenario. You're about to go to bed. You want your Codeex agent to build a whole bunch of things while you're sleeping. Maybe you go on your phone. You kick off a task in Chat GPT's app inside Codeex. You give it a bunch of tasks. You wake up the next day. You go inside Cursor inside your plug-in for Codeex. You continue the task there. Then you go out on the go. You get a burger from McDonald's and you check up on your agents there. That's not something that was really possible before with Claude Coat. So, that's another big advantage. And what's really cool is no matter where you are, mobile, desktop, you're in GitHub, Codeex will always have context of what's going on in your project. The context carries over from mobile to desktop to GitHub, wherever you are, which is really awesome. So, it's better at determining

effort required. So, short tasks are quicker, long tasks are more accurate. It is context across all your different devices. And it is trained specifically for agentic coding task. So, it's trained to be like an employee you can trust. On top of that is a very deep integration with GitHub. A lot of really cool things you can do with GitHub here. So inside GitHub, you can actually tag codecs now in any of your repositories and you can say review my code, but you can also get super specific with it as well. So if you want to tag it and you want to say, "Hey, review security or the UI or review the database," you can give it very specific commands inside GitHub comments. And it will review those very specific things. So for me, this is powerful. I like to focus on security. So I can now go on GitHub and just tag any of my pull requests with, "Hey at Codeex, please review the security of this pull request. " And it will do that very specific thing, which is awesome. But the big question is this. Is it actually good at coding? Does it match up to Claude Code? Because listen, there have been a hundred competitors to Claude Code over the last several months that have claimed to be better than Claude Code. But at the end of the day, it couldn't build apps as well as Claude Code, including GPT5. So, I kept coming back to Claude Code no matter what. So, the question becomes this. Is it as good as Claude Code at building apps? Should you switch? Let's find out right now. So

I'm inside cursor here. I opened up a terminal with control till day. Here's how you're going to set up this new codeex and this new model. Open up your terminal. Paste in this command which I'll put down below. npm openai/codex. You hit enter on that. Really easy. Hit enter. It will install codeex with the brand new model. Once you've gotten that installed, just type in codeex and hit enter on that and you will run codeex in your terminal. Here I did this inside cursor. You can open up whatever inside whatever you want. I'm going to say allow codecs to work in this folder without asking for approval. I'm just comfortable with these AI agents at this point. Once you're in, you want to make sure you type in slashmodel. Hit enter and make sure you choose one of the GPT5 codec models. It should have it selected by default, but you can go in here. You have a diff you have a few different options. Low, medium, or high. I just choose medium. Unless it is a super complex task, then you can go high. This al also save you a little bit of money. Based on my test, medium is plenty good. And now you're in the brand new agentic coding model from OpenAI. And now we can start giving it tasks to run in this text box here. You can now use this just like you would clawed code or the cursor agent. You give it commands. You say, "Hey, can you build this? Can you change this? Can you fix this bug? This UI looks weird. Hey, I want a 3D firstperson shooter. " Whatever you want. I want a Minecraft clone. You can just type that in here now and it Codeex will start building your app out. So, let's do this. We're going to do two things. First, I'm going to have it build out an app and test it against Claude Code to see what gives better results. And then two, I'm going to give you some pro tips about how to use Codeex and get the most out of it. So, let's do the test first. Let's see how this matches up to Claude Code. So, here's the test. We're going to run Codeex against Claude Code with. I'd like to build a Nex. js project management app. It includes several tools in it, including a todo list tool, cananband tool, and full document editor, as well as a chatbot. I'd like this to be a beautiful looking app that doesn't feel like AI. I'd like for everything to be saved locally until I later add Superbase, make it a joy to use and an awesome feeling app. So, basically, I'm testing its complexity. I'm giving it a super complex prompt to see if it can build this really in-depth multi-toolled, multi-layer app. It's a full project management app with multiple tools in it. I'm going to give this exact same prompt to claude code and we're going to see how well these two can cook. So, I am going to hit enter on this and it will start working. All right. So, it built itself its own to-do list. I like that claude code works this way as well. It's going to inspect the existing project structure, which there is none. This is a brand new app. It's going to outline an app architecture. It's going to implement the UI and functionality for all the different tools. So from a user experience perspective, it seems exactly the same at least from the CLI perspective using this inside the terminal. The question becomes is how good is the result going to be and is it going to be good enough where I'm willing to use all the new switch my user experience to the mobile app and everything else Codeex has to offer, right? Because if the code output's not good, I'm not going to create this whole new workflow for myself where I'm passing off from the mobile app to the plugin to GitHub and all that. So let's see how this goes. So, I also kicked off this task as well with claude code on the side. So, we'll see how that's going in a second as well. It's writing all the code. You can see all the code over here if you're using this along with me. Pretty standard stuff, exactly like the Claude Code experience. All right, so we're about 15 minutes in here and it looks like both codeex finished as well as Claude Code. So, let's run both of these and see how well they did. Just comparing the outputs from both tools. I think I lean the Claw Code output a little bit more. So looking at chat GPT here, the new model, uh it basically says it's basically talking to itself. I'm noting the test. I'll mark the test complete. I'll compile a concise overview of the changes. Uh and then it gives me a quick list of changes which are basically one to two sentences of what it built. So it doesn't go into too much detail here. It doesn't feel as ironed out as Claude because if we look at Claude, let me pull this over. It tells me step by step exactly what it built. Your beautiful project is ready. I've successfully built it out. It gives me a whole list of exactly what it built, what it focused on. It gives me design highlights. It tells me exactly how to run it, right? Claude basically just says make sure run npm install and hit run. And that's about it. So from a user experience perspective, I still like Claude code. And maybe they'll just take time for chat GPT to iron out, but I do believe vibes are a big thing when it comes to AI. You need to have good vibes. and Claude since day one, since sonnet three, uh I think have had the best vibe. So, it is no surprise that Claude Code is winning the vibe test here. Just more descriptive and feels like you're talking to a human being. But let's check the apps out. That's all that matters. So, starting

with the Claude Code project manager, it is called Project Hub. One thing I asked the AI was, "Hey, please don't make it look like an AI app. " Uh, honestly, this still kind of looks like an AI app. Basically, anytime I see blue or purple inside an app, I'm going to say it looks like an AI app. Blue and purple is just the signature of an AI builder. For some reason, they love putting blue and purple in every app. So, here we go. Project hub. It shows us, okay, we can go look at the to-do list. Let's Here we go in there. We can say use codeex. I can click uh add task. And it gives me the new task. Oh, that's for the search. Okay. So, use codeex. Add task. Boom. There it is. So, it has a good looking uh task manager here. That's nice. Looking at the cananban board to-do in progress review. So, we can say testing out codeex is in progress. Add adds the card. Looks nice. I can move it across. That's nice. A little click and drag. So, that looks good. It has an AI chat. So, I'm going to assume this is not hooked up to AI in any sort of way. So, it's probably just a shell of a chat, but it looks good. Go to documents. And that is broken. So, the documents, we have one tool that is broken, but overall the tool looks pretty good. And I'm sure I can fix this one. Hey, the documents is broken. Let's go into Codeex. Let's see what Codeex has for us, though. Aurora Studio. Of course, the blue and the purple gradient. So, just like Claude Code, it looks exactly like an AI app. I'd say it even looks more like an AI app than Claude Code from a design perspective because it is just blue and purple out the wazoo. Everything is blue and purple. when they train these AI models, they say, "Hey, make everything blue and purple. " Why is by default every AI built app blue and purple? I don't get it. Okay, so it looks like it's just one single page for all these tools. Uh, so tasks. Here we go. Tasks. Test codeex. Let's add the task here. Boom. Has the task. If I click it off, does it check it off? Boom. Look, looks good. The task list looks good. Momentum board. So, this is the Cananban board. So, let's see if we can add a card. uh test codeex add the card. There it is. Can we move it around boards? No, we cannot. Oh, we click that button that moves. So, we can't click and drag it. Didn't add the click and drag by default. So, that is not great, but it still looks good. Still looks solid. Living document. So, it has a document editor. So, we can go in here weekly strategy codeex. That works. That's I mean that's better than clawed code that didn't work here before. Uh and then lumen chat reflect with an ambient teammate during the chat. Hello, send. Okay, the chat looks like it works. That's pretty cool. Um, at the end of the day, from a design perspective, I far prefer Claude Code's output from a look perspective, from the fact that for some reason they put this all on one page. I like that Claude Code kind of split it up into different tabs and made it much more like a toolbox. The canban board was way better in Claude Code because I so I can click and drag. But at the end of the day, this is still solid and this is still way better than GPT5's output, right? They clearly upgraded the coding capabilities of this model. If I'm going to be honest with you, I did not like GPT5 whatsoever for coding. Like whatsoever. Most of the inputs I got frustrated me were not accurate, were not what I asked for, were broken, were buggy. A lot of people love it. I don't know if I'm just cursed or something when it come came to GPT5, but for coding perspective, I just didn't like it at all. But this is good. I mean, this is way better than GPT5. For me though, from a pure coding perspective, I still like Claude better than what it gave me. Obviously, this is one test. We could do a 100 different tests here, but it's consistent with what I'm seeing with any test against Claude Code. I just like the Claude codes output better. Now, let's talk about the user experience perspective really quick. Again, this has handoff, which means you can go on codecs on the web, your mobile device, and you can kick off tasks wherever you are and then come back to your IDE to do development and editing, which is really amazing and something cla uh this is actually better, right? Again, you can try this in your IDE. You can do it in the terminal. You can use codecs a million different ways, which is something that Claude Code doesn't have. So, that is a win for codecs. But I'll be honest with you, at the end of the day, what I care about more than anything is coding quality. If it can't write good code, if it doesn't have good output, if it doesn't feel like it has good taste, none of the user experience stuff or anything else matters. All I care about is output and that's it. Right? If the output is the best, then fine. I'll do the mobile and I'll do all that. That's great. But at the end of the day, my number one factor by far is the output. And just based on some quick tests on my part, I still like the output Claude gives me a little bit better. But I will be playing around with codecs and trying some other things to see what I can get from a power tips perspective. Here's a few things I'll go over with you. Number one, you want to use your model very strategically. If you're doing quick bug fixes, type SLmodel and use GPT5 codeex low. If you're doing some massive big changes, right, you're completely refactoring parts of your code. I would use codeex high and I'd be very strategic about which I use just to get better results with better timing. So, one, manage your model. Number two, make sure you have an agents. md. This is going to be the rules file for codecs. You want to have a rules file here. I I use with basically all of my AI agents, whether it's codecs or claude or anything like that. I will take my rules file I use and paste it down below in the description. So, make sure you take that and put that in your agent. mmd file. And then the last tip I'll give is make sure you use GitHub with all your repositories. This is obviously be obvious for people who do a ton of coding already, but if you're new, put all your code in GitHub because then you can use Codeex on the web and in your mobile app to actually kick off tasks on the go. So if you're in your GitHub app on mobile, you'll be able to very easily go into Codeex because it connects with your GitHub and kick off tasks on the go. So you can take advantage of the handoff between the different platforms, which is really cool. That is OpenAI's new model, Codeex GPT5. It is really, really amazing and worth trying out. It is, in my opinion, the closest to Clawude Code when it comes to coding of any model I used. It is basically just as good with some really interesting UX improvements with the handoff. Give it a try. I'll put the link for it down below. If you learned anything at all, make sure to subscribe and turn on notifications. All I do is create awesome videos on AI. and I'll see you in the next video.