LLAMA 4 Coder: FULLY FREE AI Coder! Fast + 10 Million Context!

Meta AI is finally back as today they've launched three new models. Firstly, we have Llama for Scout. I know on the main channel we had a mishap, but we had this beautiful new model which is a 17 billion active parameter model with 16 experts and it has a record-breaking 10 million token context window which is just insane cuz it can practically take out rag with this model. It outperforms Gemma 3, Gemini 2. 0 O flashlight as well as Mistral 3. 1 across the board. Then secondly, we have Llama 4 Maverick. This is the same 17 billion active parameter model, but in this case, it has 128 experts, and it's a beast at image grounding, which is even better than GPT4 Omni and Gemini 2. 0 Flash in key benchmarks. It matches DeepSseek V3 in reasoning as well as coding with half the parameters and it has an insane performance to cost ratio which is scoring a400 on Alam Marina. Now on Llama 4 Behemoth which is the third model that they've released. This is something that isn't actually available yet. It's still in training but it already outperforms GPT 4. 5. You have Claw 3. 7 Sonnet that outperforms as well as Gemini 2. 0 Pro on stemheavy task. This is a powerhouse that distilled the other two and it's interesting to see why they didn't actually compare it to Gemini 2. 5 Pro yet. Now, I personally tested this model out on the main channel and I highly recommend that you take a look at that video. But in regards to coding, I've noticed that this model isn't super great at it. It's pretty decent. It goes up against the same quality as Deepseek V3. And in terms of like the outputs, I would compare it as an alternative to Gemini 2. 0 Flash, which is a great alternative. I would say it still has a lot of bugs and there still isn't uh the same sort of generation that you would see like Claude 3. 7 Sonnet or with Gemini 2. 5 Pro. But then again, these models are costefficient, meaning they are pretty inexpensive to actually use, which is why many would actually use it. And if you are wanting to use longer context, you would want to use the scout cuz this is a better way for you to process larger code contents like your code base as well as large documents. Just take a look at this demo video which showcases that this model might not be great for coding. Now, obviously, this is just one test, but the Scout as well as the Maverick underperformed the 40 as well as Gemini Flash, Gro 3, Deepseek V3, and Sonnet 3. 5 as well as a 7 on the Kcore benchmark. Now, guys, I know I always state the LM Arena thing, but this is kind of a misconception that many people uh don't actually realize. The ELO maxing on LM Arena doesn't actually create the best model. So if it's ranked number one, it doesn't mean it's actually the best coding model. I will say one good thing about this model though, it is rapid in terms of the output. It could generate code for you instantly. It could process text as well as images, different modalities quite quickly, which is why you might would want to use something like this cuz it's a great costefficient alternative to many of the other models we see. And this is why I said it's a great alternative to Gemini 2. 0 Flash. I personally would use the Gemini 2. 0 Flash overall as a model versus the Llama 4. But in terms of maybe using it for its longer context with the Scout model, I would obviously use the Meta models over Gemini. But in this case, I will tell you that the response speed is definitely a great additive for this model, which is why you would want to use it for coding plus the longer context. But regardless of what I have said about the model, what we're going to be doing is testing it out through Klein. And this is where we're going to be combining the capabilities of the large English model with this autonomous AI coding agent that you can access directly within your IDE. This is something that's capable of creating, editing files, executing commands, browsing the web, and so much more autonomously. And remember guys, all this is completely free. client is a free extension and you can get the free Llama 4 API from Open Router. To do that, just head over to Open Router, create an account and just go ahead and search up the free scout and the Maverick APIs. And then you can go over to the API section and create a new API key that you can then later paste into client. I personally have client installed within Visual Studio Code, but you can have it operated in other IDs. Make sure you go ahead and install your ID first and then you can go and install the client extension through the Visual Studio marketplace which is the marketplace for uh VS Code. Once you have it installed, click on this and then you can open it up within VS Code and you'll be able to then access it on the leftand panel which you see over here. So this is where you can then go ahead and start generating anything

within this chat interface. Click on the settings tab. Go over to open router. Paste in your open router key. And then I want you to go ahead and select the meta model. Once you have found the model that you want to work with, make sure it's the meta 4 model or why did I even say meta? But make sure you search up the llama 4 model and you will be able to see the Maverick free model as well as the free scout model. So let's get started. We're going to first test out the scout model and we're going to go ahead and send in this prompt. Now I want you to take a look at how fast this model is. You can see that it is going to go rapidly working on generating the code. Now all of these breaks is because it is processing it through client but you can see how fast it is in terms of generating the code, executing these terminal commands, writing the code and you will see that as we go further into the video but overall I am impressed by the speeds of this meta for scout as well as Maverick model. Now, there we go. We have the first iteration for a SAS website, and I told you it's not the best coding model, but it could generate code pretty quickly. Now, this is the scout model, but let's see what the Maverick model is capable of doing. So, now I've changed the model to Maverick, and I've copied the same prompt, and we're going to send it in. And let's see what this Maverick model is capable of generating. And there we go. I'm not too impressed with this, but this was the uh output that I got from the Maverick model and this was what I got from the scout model. Now, in my initial test though, I did get a better generation for a sticky note app and it was actually functional, but this is definitely not impressive at all. And this is why I said it wasn't too impressive in terms of coding capabilities. Before we get started, I just want to mention that you should definitely go ahead and subscribe to the World of AI newsletter. I'm constantly posting different newsletters on a weekly basis. So, this is where you can easily get up-to-date knowledge about what is happening in the AI space. So, definitely go ahead and subscribe as this is completely for free. Next up, I'm going to have it create a task list management app. Now, for some reason, I don't know why my uh actual client extension isn't working, but I'm using Abacus AI chat LM. And in this case, I got this task list management app. It's really basic and I'm able to add tasks like create a YouTube video and then we can go ahead and click add task. So I guess this is functional but let's see if we can iterate it even further. Let's just say add more features and make it look way better. So let's see if it's capable of iterating on this even further and improving the design. Now, I don't know why, like I said, I'm getting some sort of error where the API request keeps on putting out tool uses and there we go. We actually have a pretty decent taskless app, but it's still pretty basic and it's not something too impressive. So, you can see that the capabilities for coding, it doesn't go more creative than what you're told. It just goes ahead and generates the bare minimum. Obviously, it's pretty decent and with its size, I guess it's okay. And in speed, but overall, I wouldn't use this for coding. I guess it's a great alternative. But still, maybe for textual prompts as well as working with different modalities. This could be a valable option to use this model. But overall, I definitely recommend that you try it out on your own to see if it's something that you're capable of working with. This is something that is definitely still impressive due to its context window as well as a couple of the different training methods that they use to configure this. I'm still looking forward to the Behemoth model. So definitely stay tuned for that. But with that thought, guys, I hope you enjoyed today's video. I'll leave all these links in the description below. Make sure you subscribe to the newsletter, follow me on the Patreon, follow me on Twitter, make sure you take a look at the new Llama video, and make sure you guys subscribe and share this channel around. This is a new channel where I'm going to be posting more videos like this. So, definitely stay tuned. But with that thought, guys, have an amazing day. Spread positivity and I'll see you guys fairly shortly.