Claude Code just got 10X Better (Codex + Gemini)

Added: 2026-05-06

[00:00:00]Claude Code is powerful, but when you combine it with chat GBT 5.5 and Gemini, you unlock capabilities that don't exist in Claude Code, drastically increasing its performance. And in this video, I'll show you exactly how to add these models for free and how to use them to create a far more powerful cla code which solves its greatest limitation. And if you don't know who I am, my name is Jack Roberts. I built and sold my last tech startup with over 60,000 customers. Now, I build AI startups and I show you the stuff that actually works. So, if you haven't already, go ahead, grab that coffee and let's dive straight in. So, I want to talk about the 6,000 gorilla in the room and that is Claude Codes performance. People thought they were ordering Albert Einstein and Forest Gump turned up at the doorstep. There's been a full lineage here. I'll let you click into it if you want to, but the fact is that Claude Code has recently gotten a little bit worse. Anthropic pretty much confirmed as much on the 23rd of April.

[00:00:57]So the idea here really and what we're doing now is combining claude code with the latest chat GPT model and Gemini to create something super powerful. So the idea is that one brain can have one set of blind spots. Now think about it from this point of view is that every model has unique perspectives and unique capabilities and strengths and when we combine those strengths we get better outputs. That's basically the whole thing of it. So you can think of it here if you like to that your quality drops you can get stuck and we get quality regression over time especially we've seen that with claude code and so the idea here is that we have one terminal in either anti-gravity or claude code and we leverage three different brands for three different things I know you're looking at this chat GBT thinking dude isn't chat GBTDX that we forgot about well I'm going to bring it back into this video and you're going to see how it actually makes things a lot better so think about it from this point of view is that Gemini Chat GBT and Claude have their own lens okay Now, what does this unlock? And then I'm going to show you straight away how you can actually get them in basically your chord code. Right now, first of all, let's look at this.

[00:01:56]Gemini 2.5 Pro has native video file input. I can feed it, for example, YouTube videos, and it can actually tell me and analyze it frame by frame. These are capabilities that we don't have on other models. And we can do this for up to 2 hours of video content completely now integrated into cloud code. Then you have native file input. Again, up to almost 10 hours of that. You can give 200 plus page PDFs in one go. That is pretty freaking crazy. Of course, also has a one million context window. Claw code itself is the IDE. And then we got like basically it's performance rate recalling over a million tokens. And you can just see how they're all a little bit different. But the point is they don't need to all be perfect to one thing. We can leverage the different skills and capabilities they have in various different errors. So if you look at this for example, how they stack up against themselves. I've put some data points here. The TLDDR for us though is that they have different strengths and weaknesses that we can leverage in claude code. Brilliant. So the first thing we're going to do is go ahead and install both the codeex the chative D5.5 as well as Gemini. By the way, if you want to learn more cloud code, I've got this insanely comprehensive detailed cloud code masterass coming in the community this month. So if you want to know how to build anything, design systems, compliance, turn money, this all into dollars, you can click the link down below and check that out. Cool. So let's have a quick think about this one then. So, first thing you need to do is pretty much you're going to copy this.

[00:03:12]Now, we're going to open up and we're going to need to do this in the terminal. And this works whether you're using anti-gravity or pull code app. It doesn't really make a difference. So, I'm going to use anti-gravity just to show what this looks like. And I have a fresh crisp new environment. I'm just going to pull up the terminal here.

[00:03:28]Little FYI, by the way, you can also do this in cloud code. For example, if I'm in the claude app, I can literally click on this down here. I can click on terminal and then I can use a terminal.

[00:03:35]I just think the UI is a little bit better in anti-gravity this stuff. So, I'm going to come to the top, click on terminal, click on new terminal, and then I've got this great guy here. I'm going to bring him up. Have a great time. And actually, what I'm going to do is literally enter in that prompt here, which is mpm install open air codeex. I already got mine installed, so it's already there, which is fine. If it's not, you can go ahead and update it.

[00:03:55]Then going to come back over and do the exact same thing for the Gemini CLI like so. Come back over here and then throw that one in. And that will install both of them onto your computer. Now, we're going to go ahead and verify that they exist by entering in this code here. So, we should get Codex's version to pop up and tell us what's there. And as you can see, it's come back into solders. And we do the exact same thing with Gemini once again just to verify that it's all rocking and rolling. And it is literally as easy as that, which is fantastic.

[00:04:18]Now, I've also built for you a skill in this that's going to use it for you automatically. So, whether you're coding, building apps, building websites, this will automatically go ahead and leverage Gemini or ChatGpt based on your specific use cases, which is crazy. Now, once you've done that, you want to come down ahead and log in.

[00:04:33]So literally you're going to literally copy Codex login. Come back over here, enter that into the terminal and it'll open up a new page. You're going to go ahead and sign in to your chat GBT account and it will authorize it and then you'll be fully connected and you can use chatg. And then once you've done that, you can come back over, copy Gemini and do the exact same thing. For example, in your terminal, you hit in Gemini and then you can literally go ahead and sign in and you just connect your Gmail account and now we can use that together. Wonderful. So now we have those installed onto our computer. The next thing we need to do is get two plugins, one for Gemini and one for chat GBT that will connect it to Claude Code, which then basically means that we can access those within any of our Claude code sessions. So basically what they do is you have a Codeex plugin and a Gemini plugin. This one lets you do things like review. So you could create something and it will actually review all of your output. It has adversar review which is basically like targeted to be like as critical as possible. Kind of like playing devil's advocate. You can basically rescue, you can review. Then Gemini is really cool because we can just summon it and do various different things. Ultimately though, all we're doing with these two things, these two GitHub repos right here, this one here is for the Codeex plugin, which is really awesome. You can see it's piping a popping. And then we got the second one, which is for Gemini package, which is fantastic. So, all you're going to do is come back over to the notion document. You're just going to copy and paste them in one at a time by clicking copy. Coming back over to anti-gravity.

[00:05:51]And by the way, if this all sounds like I'm speaking Spanish, you don't. This means watch this master class right here. I break down exactly how to use anti-gravity, but get you fully up to speed. So again, we just come in, we enter in the and basically just one by one, we go ahead and we just add them all in. And then within anti-gravity, again, you're just going to open up Claude one more time by typing in Claude and then just paste everything in. So there you go. First one, plug-in marketplace. It'll add it in. You come back over, you come down, you copy the next one. You do the same thing. Come back over here and just follow that through step by step again. And then you repeat that for step three, step four.

[00:06:22]And when that's all done, come down and get reload plugins. Why? Why am I doing that? Well, basically it's cuz it refreshes and then you can actually access it within the environment. So now that's done. Let's give it an example test. Okay. So I'm going to say to Claude something really simple like, "Hey, what I'd like to do for me is write for me a 1,00 character guide on the most effective way to build AI SASS in 2026." So let's go ahead and actually show you the capabilities that this unlocks. So I'm going to go ahead and open up a new terminal. We're going to bring up Claudia for example. Now, the beauty of this is that you're using your existing chat GPT subscription. So, if you're on that $20 a month, there is no additional cost for you doing this. And I'll show you exactly just how much you get in a second. But, let's give it a command. Now, this would work really well with I mean, this is a random thing. I'm asking it to write a JBT or middleware for Express, reads a token from blah blah blah, verifies it, all this cool stuff. I'm going to say, "Hey, do me a favor. Save this for me in a folder on the left hand side." Cool. So, it's gone ahead and it's actually written that for us. Now, all you're going to say is, "Awesome. I'd like you to use codeex to review your work. Um, everything you've just done for me, plain language. Send that one over there and send it off." And the cool thing about this, again, it's no additional cost. It's just using your existing chat GBT subscription. Now, there's a lot of people here who are saying that they're leaving Claude code entirely and going over to Codex. Now, personally, I have noticed a bit of a drop in performance with Claude. I've asked it to do things and I've sat there thinking like this is just not correct. It's just like absolutely not right. So, Cloud Code has definitely skipped a beat from like a month or two ago where it felt kind of like it was uh magical. Now, what does that mean? Does that mean that we drop Cloud Code and run over there? Well, just like you, I have no allegiance to any particular model. I just want to use whatever the best thing is. Now, Claude is actively working on these things in the background. I don't think that it's time right now to just completely never use Claude code again. But what I do think is a really huge unlock is leveraging codeex check GBT 5.5 Gemini to really uplift and check everything you're doing include code and if it's at the point where it's just like a thousand times better then we simply move over. So I wanted to call that out because let's just keep it 100% real.

[00:08:28]That's what we want to do here. And just like that codeex has gone ahead and found real issues. It's summary of what it's worth acting on. No JW secret algorithm. It has given us basically a high medium and low breakdown of everything that's happened. So this literally this is exactly how this works. You get think about it like this.

[00:08:43]Okay, Claude code is the one creating the actual code and then we can bring in tag in codec as a reviewer to make sure that it's not actually hallucinating and then we can go ahead and try out Gemini which can actually look at videos. So for example, here's a video of me. I'll say a fake news fake news video of me actually of me just some say this isn't real. Although I will say this that bottle does not exist anymore. So let's say that we've got this video and then we want to do is get Gemini to analyze it for us. By the way again no cost.

[00:09:09]Well, at all. So, let's open a new window. Uh, and we'll come ahead and just do Claude. And I'm going to say something like, "Hey, that I'd like you to use Gemini to go ahead and analyze this video just to ask it to describe what happens in the video and just give me a bit of a breakdown. Then save it for me on the left hand side as video analysis." Okay. And then I'm going to give the file save file name is boom.

[00:09:28]And just drop that in like that and send it off into its journey. And just like that, we've got our feedback. So, a quick recap of what Gemini saw. A man in a teal hoodie in cap holding a white Stanley water bottle in the studio.

[00:09:38]large green gold clock pro mic over four seconds. He crushes the bottle bare-handed and he ends it with a dead pan look at at the crumpled result. No dialogue, no music, no on-creen text, just ambient crushing sounds, comedic anticlimatic vibe production test style.

[00:09:53]So why is this cool? Well, now it means that we can just leverage Gemini to analyze videos. So maybe you think about these long audio files, huge PDFs that are going to be so difficult to do and claude, we can bring in and tag in Gemini now to solve those things for us.

[00:10:06]Now, tagging these models in is one thing, but in reality, when we're drinking our beautiful coffees and we're having a good time, we want Claude to tag in the best model for the job without us even thinking about it. And we can do that using a skill that I made that'll be available for you down below completely for free. We tag it in and then we can literally accelerate the performance of Claude Code. Now, before we download this skill, I just want to give you a full touch on what this actually costs. So, in Claude, if we're using Claude Code, we've got our either $20 a month subscription or our $200 subscription. The chat GPT is fantastic, by the way, because that's only 20 bucks. But what that gives you practically is around 50 GPT 5.5 messages every three hours. I think that's super reasonable. On the pro tier, it's effectively unlimited GBT 5.5 on the pro mode. Uh codec author gets the same sub, so no extra per token cost. You don't have to use any of your API keys. Plus is enough for daily code reviews and rescue calls. Pro is for full days of background agent work. So you can sign up based on your use case there. And then we've got Gemini. So you get 1,500 flash requests a day. That's for quick text and image jobs. You get 50 pro requests a day. So these are your big video, your audios, your uh big PDFs. For most people watching, that's going to be absolutely fine. You're not going to need to upgrade. You can just literally keep it at $0%. Then each request is one full prompt. Could be two-hour video or a 200page PDF.

[00:11:26]Incredibly generous from a Gemini team.

[00:11:28]I think that absolutely crushes it. I honestly this setup is so freaking sick.

[00:11:32]I don't know why more people aren't using it right now. So, what's our strategy? Well, our strategy is to combine them together into one really cool thing into the same bill, which is really freaking cool. So, don't take my word for it. You can run the math. I'll let you look at the numbers. It's super duper cool. But what we're going to do now is actually bring these together with the skull to show you just how powerful it is. And this is what I call the three brain auto router skill. So, idea is it's one prompt, three brains, and it brings out the right one for the job. If it's video, audio PDF, we're going to use Gemini. If it's build the refactoring claude, and then GPT is the review. And remember, GPT just sits there looking at the code, making sure it's perfect. Now, to do this, just click the link down below in the description. You're going to come down and you're going to go ahead and grab this three brain skill. You're going to grab it like so, and you're going to download that either with an anti-gravity or claude or your integrated development environment of choice. Just fancy word of saying where we do the coding. Just go ahead and drop that in there. So, I've dropped in my three brain skill, which is very, very cool. It's got loads of interesting details. They're going to say the below which is, "Hey there, I've just added in a threebrain skill MD. Familiarize yourself with this and add it to your skills." And just like that, it's got everything. And I've just zoomed in just so you can get a good full look of what it's done. So here's the summary of this gun. So you got Claude, which is him, the building driver, the IDE harness always in front. Codex is the reviewer, the second brain, and the rescue. Then Gemini is the eyes and the long contact.

[00:12:47]So key rules that I've internalized, no self-re law. So here's the big problem with Claude is that when it gets something wrong, it doesn't recognize that it's actually getting it wrong. So you can sit there and it's stuck in like a roundabout loop of keeps it overthinks. So what we do with this is add in a hard stop rule. That basically means if it's failed to solve it once, we're going to tag in just like a WW rest in the match. We're going to tag in somebody else to come and save the day.

[00:13:15]That's when GPT enters a situation.

[00:13:17]Failure detection. Again, if I fill the simulation twice in this case, I must hand off the codeex review automatically. Risk path detection, edit touches. We got Gemini roots for video, audio, PDFs, parallel consensus. When you say ask all three, all models answer the same question in structured format and he differs out the results and then output filing and stay asleep rules.

[00:13:35]Cool. So, let's actually put this to the test in one particular thing. So, I'm going to give it a really cool product.

[00:13:39]Hey, there in my downloads is a video from Alexi. I want you to analyze the entire video. I want you to tell me any of the text that appears in the video because the audio is one thing. So, I want to get the transcript, but then I also want the visual overlay. I want to take both of those components and I want to take that and create a beautiful HTML document uh that I can look through and learn from interactively, combining both what's said and what is seen visually.

[00:14:02]And when that's all done, I just want to make sure that Codeex reviews that to make sure there's no mistakes, that it's uh accurate and completely on brief.

[00:14:09]Thank you. Beautiful. And then we send them one off and let it work its magic.

[00:14:12]And guys, look at this. It's running both tools. Codeca, the Gemini CLA. Now, let me set up the output directory and start the Gemini pre-processing pipeline. Preparing the video and extracting audio for the transcript. The video is 8 minutes and 55 seconds. Since it's under 10 minutes, I'll send the full video to Gemini, but downscale it to 720 pixels to stay within limits.

[00:14:32]Freaking cool. Now, it's extracted the audio. It's doing full all this sort of stuff for us, and we can send that one off. And just so you know what the video looks like, this one's Alex. I saw Alex last week in LA for the school games which we won which is really cool. He is uh pipping and popping hitting the gym and doing various different things. So cool. We get a visual overview cuz you know it's like you think about it like this right? Maybe you're looking at various different videos. There's a bazillion use cases for this but the idea is like you not you would never get this on any regular analysis and it's there for you when you need it or if you've got like a huge PDF for a client like massive contract review. We can't realistically just depend on Claude Opus 4.7, right? We want to tag in the right experts that actually make sense at the right times. Now, as you can see, guys, this is all coming quite nicely. Look, it's got a visual analysis. Wide shot of Alexi seated at a wooden desk. The background consists of three black vertical panels separated by white light strips. And guys, think about this. If you want to grow on Instagram, you want to grow on Tik Tok, what you say is one thing, but the visual aspect is something completely different. So maybe I want to reverse engineer why certain people are going viral. Well, I would 100 bazillion gazillion%, that's a real number, use this because I can do a visual breakdown and I can do a transcript breakdown. If you just look at the words, you're leaving almost all the well, not all the value, but you're leaving quite a lot of value on the table. This lets us do all this. It's now going ahead and build the HTML document. Then we can see exactly what we're dealing with. And then guys, we've got this full video analysis that has almost line by line and then a visual overlay. Look at this wide shot of Alexi a wooden desk. Two books, one that says $100 million leads, one that says $100 million offers. Have you noticed, by the way, a trend of those appearing in YouTube videos? I think I actually spoke to him last week. I was like, "Dude, did you know people put your books at the back of their videos? Was so freaking funny." But look at this. You got all this different stuff here. You got visual over overlays. Oh my gosh, it'd be great if I could talk today. All this cool stuff. And when there's something that changes, a large sheet of white paper covers the desk. On the left hand side, a handwritten blue mark status, power, credibility. wouldn't pay a single dollar bill for any of this visual overlay analysis that just went all the way through Gemini. We didn't have to do anything. That was called automatically, which is awesome. Then guys, you can keep on scrolling down here in the bottom, we get nice codeex reviews of findings. But again, Codeex is best place reviewing code, which is really incredible doing like I showed you earlier. But this whole thing came together with literally one prompt. And in addition to that guys, I can also give him a prompt like this that just says, "Hey, I want you to use three different models. Tryd 5.5, Claude, and Gemini. debate. What is the best social media platform to start a business off in 2026? Now, I can pull in these models for thinking questions. Obviously, Claude can create sub aents to debate things, which honestly is one of the most underrated hacks that you can do.

[00:17:08]Literally, in claude code, you can say something like, "Hey, look, I want you to debate these things. Create five sub agents. Five is about the quality threshold maximum after which for some reason there's just some diminishing returns principle. I think it's to do with the main claent's ability to dip and dive. But here we can bring in these different models to essentially debate different things for us to give us a better response. And then just like that we get the different perspectives at the bottom. They they debated and then we get the verdict which is YouTube winning long-term a platform where content compounds like an asset combining search discovery and audience ownership. This drastically improves Claude's performance. But to get the most out of Claude, you need to know how to build super skills properly, which we're going to learn by watching this video right