Llama 4: DESTROYS ChatGPT & DeepSeek? 🤯

So, Llama 4 has just been announced and according to LM Arena, it is beating chatbt40, Grock, Deepseek R1, and Gemini 2. 0 along with pretty much every other model apart from Gemini 2. 5 Pro. So, what we're going to be doing today is testing it against these different models. Now, if you're not sure, okay, what is Llama for Maverick? What does it do, etc.? So it's got a 10 million token context. It is free to use. So you can actually get a free API directly from open router or you can use it directly inside LM arena for free as well. And

basically Llama for have just released four different series. All right. So Meta is behind this, Mark Zuckerberg, and they've released three different models, which are Llama for Scout, Llama for Maverick, and a preview version of Llama for Bearmouth, which is still in training. Okay, so if you're wondering, okay, what are the differences between these models? You can see some examples here. So multiodality, mixture of experts, architecture, long context support. So, even Llama 4 Scout, which is a lightweight model, now supports a context window of up to 10 million tokens, which is pretty insane cuz the last one, the Llama 3 update, actually had 128 tokens. All right, and you can see the comparisons right here in terms of parameters, total parameters, etc. And also, this is completely free to use on Open Router. All right, so we're going to get into straight testing

these different models. If you want to see the leaderboard on LM Arena, you can actually go onto LMA arena. ai and then you can see the models right here and you can see that Llama Maverick 4 is outperforming CHP40, Gro 3, Gro 4. 5 preview, Gemini 2. 0, Deep Seek, and Deep Seek 1. So, what we're going to do is side by side, we're going to test these different models. So, what we'll do is we'll put Llama 4 over on the left hand side. So, you can see Llama 4 there. and then we can compare this to some of the best models out there and see okay how good is it actually right all right so

first thing that we're going to test is how does it perform versus claude 3. 7 and we'll use this prompt which is create an AI powered audit tool for goldie agency that analyzes a business's operations and suggests automation opportunities in HTML users must enter their details etc right in HTML format so now what we're going to do is hit enter and we'll see how these perform side by side. So you can see here that Clawude 3. 7 Sonic gets straight into coding it whereas Llama 4 has a little think about how it's going to work first and then starts coding this out. Does seem like Claude 3. 7 Sonet is a lot faster. But we'll test these out and see how they perform in a second. This is coding the CSS separately. All right. So claw 3. 7 Sonic is creating everything in one single HTML file whereas for example llama 4 has created a separate CSS and HTML. So let's see how they perform. We're going to plug in a new liveweave mode right here and we'll plug in the HTML from Llama 4 on the left hand side. And then we'll grab the CSS as well. So you got the style CSS here. Plug that in there. And then we have the JS over here. Right. So if we have a look at the output, the form is not bad side by side. It's pretty bland in terms of the design and stuff like that. Let's see how Claw 3. 7 Sonic performs versus this. So we've got the output from Claw 3. 7 Sonet over here. That's coded oneshot HTML. We'll plug that in. See how it performs. And there's not even a button to use it. All right. So, if we compare side by side, which one has performed the best here, I would take Llama 4's output simply because the form actually works, right? It's actually got analyze my business section here. You can, for example, select between the industry, put your business name, number of employees, etc. Whereas, for example, the design is slightly nicer, but you can see here there's actually no button to get side right there. Honestly, I'm not impressed with either inputs, but I would say Llama 4 actually wins on that particular model. Right. Next up, what we're going to do is we're going to test uh reasoning example. So, I'm going to take

this challenge right here and say there's a tree on the other side of a river in winter. How can I pick an apple? We'll plug that into the models right here. And this time we're going to select Gro. All right. So we have Grog 3 preview over here versus Llama 4 and we'll see which one performs the best. If we actually have a look, the output from Llama 4 is a lot nicer formatted. It's easier to read, etc. They both give quite good solutions side by side. But we'll wait for them to reply and then see what we get here. So I actually prefer the output from Llama 4 versus Gro 3 Preview. So if you look at Gro 3 Preview, it's like long water text. It's not very nicely formatted. And additionally, side by side here, you can see that this I this finds a constraint straight away. All right, so the tree is on the other side. It's winter, so the tree probably won't have any apples and we're questioning the assumptions here. So, it's actually figuring out, okay, there's a lot of problems with this question. And then it comes back with some ideas in the framework. All right. So, I would say the answer from Llama for Maverick is much better than Gro 3 side by side simply because it's nicer formatted. It focuses on more solutions and it identifies the problem straight away. So, the constraints of the problem. All right. Next up, what we're

going to try and do here is create a sale plane snake game classic using HTML. We'll compare this side by side versus chat GPT. In fact, no, we'll use Deep Seek one for this. All right. So, we'll compare this side by side versus deep sea car one. So, we got llama for Maverick over here versus deep sea car one. We're going to say create a selfplay snake game using HTML of a simple GUI. And we're just going to say make it one single HTML file and we'll see what we get back here. They're thinking about it. Looks like it might be broken. Let's give it another go. It's a false start, my friends. We're just going to refresh the page there and try again. And LM Arena's totally broken on us. So, what we're going to do is go over to open router instead. We'll start a new room here. I'm going to edit this and we'll add two different models here. So, I'm going to test Llama Maverick and we'll test that side by side versus Deep Sea Car 1. We'll use the same prompt here and see what we get back. So, one thing to note here is like when you're testing them out, it looks like Llama 4 is more of a base model. So, it doesn't really have that like thinking capability. Whereas, for example, if you look at Deepc1, it's going to reason first before it builds out the tool, which I think is much better for code. But let's see what we get back in a second. We've already got the output back from Llama for Maverick over here. So, we'll copy that and we'll plug it into the HTML. See what we get back. Refresh the page. And that is no bueno. I tell you that for free. That's cuz it's create it's not created one HTML file like we asked it to. It's created three different ones. All right. So, we're just going to refresh the page here. We'll take the code. So, it didn't listen to my instructions. We just create one single HTML. But we'll grab the CSS and test this out anyway. See if this actually works. And then we have the JS over here. So, let's plug that into LiveWave. That is fast, isn't it? That is one of the fastest snake games I've ever seen in my life. It's incredible. To be fair, it works perfectly. So, I can't complain there. All right, let's see what we got back from DeepCar 1. So, it's created a one single HTML file, which I like. That's what we prefer. And then we'll grab this and we'll plug it into something like LiveWeave again. Test this out side by side. All right. And Deep Sea G1 has totally failed on that. Look at that. It keeps failing. I honestly I think I prefer Grock. Look at that. But that's still going. It's playing itself. It's having a great time. If we have a look at DeepCan, it's not bad with the buttons here, but it kept breaking before if you saw that. So, you see here how it keeps breaking. I'm actually going to go with I would say the Llama 4 beats Deep Seagar one there. It's created a better output. It's faster. It doesn't have the UI buttons, but if it actually works and just carries on playing itself, then that's much better for me. I'm going to take that output any day of the week. So you can see here for example it just stop and it's not working at all. All right. It's not even counted the score as well. So you see here how it says score zero when it should be six or four at least. There we go. All right. Ella Marino actually finally came back to us which is interesting too. So last thing that

we're going to test now is creating 3JS games. So, we'll grab a prompt from the AI profit boardroom. We're going to say, "Make me captivate an endless runner game. Key instructions on screen. Use p5. js, no HTML, pixelated down, and interesting backgrounds. " And what we're going to do for this is we're going to use Gemini versus uh Llama Maverick 4. Right. So, if we go to Llama Maverick 4, we'll use the paid version just to make sure it actually works. Let's plug that in. And then we'll do the same on 2. 5 Pro. We'll do that inside AI Studio. And then if we go to create prompt 2. 5 Pro over here and we'll run the same prompt through both of these. All right. So let's see. Llama 4 has come back to us with the JavaScript. Let's check that out. Seems to be remarkably fast. We'll see how that performs. This is creating the look at that. It's not working. My I'm really not impressed with Llama 4. It totally broke. It seems to be slightly drunk and it's saying it's game over with a final score that keeps going up even after the game was finished. All right, so I would say by default, Open Router has totally failed here. Let's see what we get back from GMI 2. 5 Pro, which is still having a little cheeky think about it, building it out, and then it should come back to us with an answer in a sec. We'll wait for that to load. definitely takes a lot longer than Llama for Maverick, but if the outputs are better, it's worth the wait, right? You can't rush greatness, especially when it comes to pixelated dinosaur games. So, let's grab the p5 js here. And if you want to preview this stuff, just go to editor. p5js. org. Plug that in. We'll hit play. And boom. Yeah, it's just the quality is just 10 times better. Much better, isn't it? you actually get an output back from Gemini. It's much better. It actually works, etc. It's in a league of its own, Gemini. And Gemini is free as well. So, I'm going to stick to Gemini and claw 3. 7 on it, honestly, from what I've seen today. But, it is exciting to see a model with a 10 million token limit. All right. Again, I don't know how it's risen to the top of the leaderboard so fast, but I always take these things with a little pinch of salt. So, thanks so much for watching.

If you want to get a community focused on making more money and saving time with AI, feel free to get that inside the AI profit boardroom. Prices are going up this month, so make sure you lock in your price now before you miss out. And additionally, this focuses on all my best automations, workflows, templates, AI agents. It's got a crash course and all my best SAPs for using AI. Top of that, it has a community of 694 members. So if you have any questions, you can post inside the community. Ask us any questions you have. Additionally, if you go inside the calendar here, there are weekly Q& A calls where you can jump on those calls, ask any questions you have, get live help. Additionally, if you want to watch back the call recordings, you can watch them back right here. So you can click on here, watch back the weekly Q& A call recordings. If you miss anything, then you can watch them back right there. Link in the comments description. Prices are going up, like I said, this month. So make sure you sign up now before you miss out. And if you want to get a free one-to-one SEO strategy session that shows you how to get more leads, more customers, more sales, and make more money from your website with SEO, feel free to get that link in the comments. We'll give you a customtailored game plan, ask any questions you have live on the call, and you'll learn how to 10x SEO traffic based on what's working for us and our happy clients like you can see right here. Feel free to give that link in the comments in description.