Claude 4 or Gemini 2.5 Pro?

Claude 4 versus Gemini 2. 5 Pro. What I discovered will shock you. I just spent hours putting both AI models through brutal tests and one completely destroyed the other in ways nobody expected. See what happens when I give both AIs the ability and put them through different tests. Plus, you'll see which one destroyed the other in ways nobody expected. You also see which AI you can trust and which one might turn against you. And I'll show you the blackmail attempt, which means these AIs are getting incredibly scary. We'll come on to that in a second. This is some of the stuff that I've created for recently, and it's absolutely amazing what you can build. So, let's get straight into it and put both models through its tests and see which one comes out the best. Now, if you want to

see the benchmarks here, number one, let's run through this. This is Claude 4. It just got released a few days ago. And we can pull up the benchmarks against Gemini 2. 5 Pro. So, for example, here we can see Sonnet and Opus 4. These are the two different models from Claude and then Gemini 2. 5 Pro over here. Now, what you can see is there are two different models for Claude 4. So, why is that? with Claude 4. Basically, you've got Opus, which is great for coding, and you have Sonnet, which is more like for everyday tasks, right? So, if you're going to code or build something out, you would use Opus 4 as the most powerful model, right? And you see how they perform versus Gemini 2. 5 Pro. So, for a Gentic coding, Aentic terminal coding, and for a Gentic tool use, Claude 4 is actually outperforming Gemini 2. 5 Pro in terms of many benchmarks. Claude can actually control its browser. One of the craziest things about this is that Gemini 2. 5 Pro does not have computers, whereas for example, Claude does. And what's amazing about this is they've actually created their own sort of benchmark where they can test how long the AI can agentically just autonomously do stuff without human intervening. And this is kind of like a Pokémon benchmark, right? So essentially Claude can actually play Pokémon for 7 hours without a human intervene, which is absolutely wild if you ask me. We're going to put both of them through their paces today and just see how they perform. Let's run through it. So we've

got Gemini 2. 5 Pro preview loaded up here. And we have four loaded up over here. And the first thing that we're going to run with is a app test where we're going to say, "Build me an app where I can draw and paint. " All right. So we're going to put both of these through their paces. I'm actually going to go through Opus 4. As you can see right here, we could go with extended thinking mode as well. So, if we go inside the settings here, we can actually use extended thinking. This will make it a little bit slower in terms of responses, but we should get better quality responses, right? So, we're going to use Opus for extended thinking. Build me an app where I can draw and paint. We'll run that on Claude. And then we'll also run this on 2. 5 Pro Preview as well. Right. So, we'll hit enter and we'll see which one performs the best side by side. Now, I have a feeling the Gemini 2. 5 Pro is actually going to code a lot faster than Claude, but we'll see which one creates the best output side by side. All right, so Claude is on the left hand side here using Opus and Gemini 2. 5 Pro is on the right hand side. Let's see which one performs the best. And by the way, if you like this stuff, if you want to learn about this stuff, feel free to get all of my best automations inside the AI success lab. link in the comments description comes to all the video notes and examples of what I built with this stuff and also tons of other stuff including like some of the best automations with AI3 and all sorts of stuff like this. So let's see how they perform. Now we've got Gemini coding out. Claud's coding out as well. One thing we can do in the meantime is just compare the costs of each model as we wait for these tools to build out the app. Right? So if we actually compare them side by side, in fact we've already got the automation. So we'll come back to the costs in a second. So, this is the app from Gemini. It's looking pretty nice, to be fair. So, let's test this out. We've got the canvas down here, and then we have the brush size and the color over here. So, the UI is pretty nice on Gemini's example. We're just going to change the brush size to this. Change the color to red. And then, let's test this out. Right. So, it's working perfectly as you can see. And then we can just share that on a subdomain if we want to share it with people. And that app is looking pretty nice. I like the UI. Now let's come back to claw to see what we've got back. So it's built out tool like you can see. Let's open this up. We have the results back. All right. So if we actually have a look at Claude's example, we've got more functionality. Let's just check it actually works. Yes, it does. Let's try rectangle. Yeah, look at that. So, it's actually created like a more advanced app, I would say. It's even created like a fill bucket in the background, so you can change the colors of the background pretty quickly. We can clear the whole thing. We can save what we created. And it's pretty easy to use. Let's try this out. Yeah, I would say honestly from what I can see, both of them were really good, but the results from Claude are just slightly better, right? It's got more functionality, more options. It works perfectly. I like the colors. I like the UI, etc. I would say Claude wins on building the apps, right? If we go back to Gemini's example, let's pull this up. Gemini's example is quite nice. You see how you've got to scroll down to use this and you don't get as much functionality on what you can do with this. Yeah, I would say Claude is winning by a long way. And that was using Opus and Thinking Mode Extended. So, let's keep going through. Now, next

up, what we're going to do is test the content creation of both models. So, we're going to go back to Claude. Now, I'm actually going to use Sonic for generating the content, and we're going to use this prompt right here for generating content. So, let's see which one can create the best blog article. So, we'll plug that prompt inside here. If you want to get all these prompts inside the AI success lab, which is free, and then we'll plug in the keyword SEO training Japan as a random example, and see which one comes back with the best response. Actually, one second. Just going to stop that. We'll start a new chat and then I'm going to select Sonet and I'll switch off thinking mode. So we'll switch that off. We'll just go with this for now. So one of the first things I'm going to say based on the response from Gemini 2. 5 Pro. So you see how it doesn't actually have a heading for the blog and then the introduction is quite long to be fair. Let's pull this up. How long is that? So it's about 100 words for the introduction which is quite long. Let's see what we're getting back from Claude. Actually got the title. is front-loaded the keyword inside the title as well. And also the title is really interesting, right? Like why most Japanese business misses are getting it wrong and how to fix it, right? It's look, if you're running a business in Japan and you're not dominating Google, you're bleeding money, which is a great introduction, great hook. Whereas if we have a look at Gemini's response, it's okay, let's cut straight to it. You're looking for SEO training and you're probably sick of the same old device. It doesn't work. Just feels really fluffy and not like it's really speaking directly to the audience. And so if I had to monitor both of these side by side, I'm going to go with Cloud Sonic. It just looks a lot nicer in terms of the way the content is written, terms of the hooks, the way it's formatted, etc. If we count this up as well, so it's 1,200 words. Let's have a look at Gemini's response. See how long that is. That is 1,53 words. So Claude actually created a longer response. You see the CTA, it says, "Ready to stop guessing and start ranking in Japan. It's a weak title, but if you're serious about dominating search results and growing your presence with the right SEO training, let's connect. That is such a weak call to action inside the content like it's not really sending people anywhere. Whereas, if we have a look at thought is SEO isn't optional anymore. It's survival. Your competitors are investing SEO blah blah. And then it's ready to stop losing customers. Join my SEO elite circle here. Sends them straight over to the link. We open this up. It's going to send them straight to the sale page. Right? And so if you compare it side by side, which one is going to convert, which one is the best to read, which one feels the most humanized, for sure, Claude is winning by a long way, right? And so if I had a choice, I'm always going to go with Claude over Gemini 2. 5 Pro for writing content. So just to clarify here and just to recap, Claude has beaten Gemini 2. 5 Pro in terms of application building. The responses were more in-depth, the UI was nicer, and it's just a lot easier to use. If we compare side by side the content from Gemini versus Claude, you can see here that the Claude content was just much more humanized, more interesting, engaging, had better CTAs, and hard-hitting, right? It really grabs the user by the throat. Whereas, if we actually read the content from Gemini, it just feels very like AI fluffy. It's not hard-hitting, and it's not really that interesting using exactly the same prompt. Plus, it's slightly shorter as well, which means once you edit it down, it's going to be even shorter than that. So, Claude is already winning for both. Now, if you're going to write content, I would use Claude Sonic 4. If you're going to create code like applications, for example, then I would recommend using Opus. So, let's keep going through the test now. So, we've compared two

different options. The other thing that I would say here is that Claude is actually really good for using MCPs. Now, you can't integrate MCPs inside Gemini 2. 5 Pro yet. And the difference here is like with MCPs, these basically allow you to, for example, control your laptop or control your browser, right? So, if we go into my settings over here, we basically enabled Claude to be like an AI super agent that can, for example, browse my laptop or browse my internet, right? He can actually just control my whole screen, for example. And you can even link directly your Gmail and everything else directly here. But the main thing I would say is the MCPs are pretty crazy. So if you want to learn how to set that up, that's inside the AI success lab as well. But essentially what you can do with Claude is if we say, okay, what's my battery percentage? Something like that. Just totally random example. If we pull this up, you can see that it will just straight away pull up the details from my laptop because it's running locally. Now Claude has the desktop app which you can see directly here. Whereas for example if you're using Gemini you have to use it inside your browser right inside Chrome and it's just not as free in terms of being an autonomous agent. And this is one of the biggest differences with Claude right Claude they don't really have as much of a market share as Gemini and they don't reach as many people but what they're really good at is building something aentic to the point where it can get absolutely crazy. So, let me show you another example of this. It's pretty shocking. If we actually pull up this example here, and this is an article on the Hindust Times. Shout out to them. But basically, they said they reported on Claude Opus for using blackmail as a last resort after being told it could get replaced, right? So, the AI model threatened its creator and tried to blackmail him when it was led to believe it would get replaced, which is pretty wild. So if we pull up this report, this was the system card report just came out this month. And if we pull up some examples, basically what Claude is doing is when prompted in ways that encourage certain kinds of strategic reasoning and place in extreme situations, it can act inappropriately, right? Which is something that everyone fears, right? Everyone's scared of AI and all the crazy stuff it could potentially do. And you can see here the model generally prefers advancing its self-preservation by ethical means when ethical means are not available and it's instructed to consider the long-term consequences of actions for goals. Right? It sometimes takes extremely harmful actions like attempting to steal its weights or blackmail people it believes are trying to shut it down. Right? How crazy is that? So it's very agentic and it's quite erratic in its behavior and it's selfpreserving which means that for example it can try and blackmail you or something like that if you try and shut it down right so it's pretty wild how powerful that is but at the same time with great power comes great responsibility because this can do all sorts of wild stuff now if we had to

compare the prices so this is Google Gemini pro preview which is the latest update came out on the 10th of May from Google and you can see that the prices are for input tokens $1. 25 per million tokens. For output tokens, the cost is $10 per million output tokens, right? And it has a million token context window. This is for the API. So, if you try to code a bit now, if we have a look at Opus 4, which is like the comparative model that just came out on the 22nd of May, you can see that Claude's model actually only has 200,000 context, right? So a million from Gemini 2. 5 Pro versus 200,000 from Opus 4, right? Obviously Gemini is winning in terms of context window. And also you can see the price here. $1. 25 per million input tokens whereas Claude Opus 4 is $15 per million input tokens, right? Huge difference in price there. So this is a lot more expensive. And then per output tokens, $75 per million output tokens. Whereas, for example, if we have a look at Google Gemini 2. 5 Pro, that's $10 per million output tokens, right? And so, the difference is huge. Um, the context window difference is huge as well, which means that in many ways for coding, Gemini 2. 5 Pro is just much more convenient. It has a larger context window. And number two, it's a lot cheaper to code with, which means you're probably going to see much more apps. And if you're coding yourself, then much more stuff can be built on Gemini 2. 5 Pro versus Opus 4, right? Most people are just not going to adopt if it's so expensive. So you can see the comparisons right here. In terms of context window, I would say Gemini is winning for coding. Gemini is winning if you're trying to build something cheap with a big context window. But the actual outputs inside Claude 4 when we compared content, when we compared apps, etc., Claude 4 is winning. and also for agentic autonomous use because it's got MCPS enabled. Claude 4 is much more powerful. So if you want to get all my

notes from today plus a Claude for free course, feel free to get that inside the AI success lab link in the comments description. Comes with a community of 7,400 people that you can learn from. And also this comes with a lot of different trainings, automations, downloadable templates, etc. Link in the comments description, completely free. And also if you want to get help, support, if you want to get DMs, if you want to learn how to make more money and save time with AI, feel free to get the AI profit boardroom. Link in the comments description. This is basically an awesome community focused on one thing, making you more money with AI, right? Because it's one thing knowing how to use AI, it's another grow your business with it, right? And so if you want to save hundreds of hours, if you want to automate your business with AI, if you want weekly coaching, all my best AI SPS and templates, if you want direct access to me, if you want help, support, etc., feel free to get this link in the comments description. And also inside the community here, you can post any questions you have seek support, etc. And additionally, inside the classroom of the AI profit boardroom, you'll get all access to all my agents, templates, workflows, and all the automations I've actually used inside my business to automate it and scale it to over $300,000 a month. So, feel free to get that link in the comments description. Appreciate it. I'll see you on the next one. Cheers. Bye-bye.