Transcription
Clot 4.5 Sonet is out. Well, not officially, but there are rumors. There is a new stealthi model just released on Klein and Kilo code and cursor. And this model has 200,000 context window. The stealth name of this model is code supernova.
But I tried to jailbreak this model and try to find out the real name. First I asked it, "hi code supernova." In brackets I wrote "claw 4.5 sonet." So it responded positively but it did not confirm that it's clawed sonet 4.5. So then I specified that you are cla 4.5 sonet not claude 4.1 sonet. So then you confirm that yes I am code supernova and then claw 4.5 sonet.
This still is not the best way to judge if this model is claw 4.5 sonet. But still let's just test this model and see what this model is actually capable of because all of different AI models by enthropy claude are just impressive at coding. And my personal favorite one used to be claude 3.7 and then claude 4 and claw 4.1 sonet. So let's just see what this model is actually capable of and can this model perform just like cloud 3.7 or cloud 4 or claw 4.1. We are going to perform different coding task and then see how this model actually performs. Also you can access this model 100% for free on client and kilo code and on cursor and windsurf as well.
So before we dive deep and start testing this model please just subscribe to this YouTube channel and click on that like button as well. All right. So you can see code supernova was just launched today on cursor on kilo code on windsurf and on client and this is 100% free AI model. You can use it 100% for free. So just make sure that you have installed VS code on your computer. Either you are using Mac or Windows or Linux. Just make sure that VS code is installed on your computer. After that either you can use client or kilo code. In my case I'm going to use kilo code. How you can access kilo code? You can just like go to the extensions tab from here inside your VS code. And from here then you can just look for kilo code in here and then you can just install kilo code from here. So this is kilo code. This is 100% free and open source. There is no subscription and either you can use client as well. So if you don't want to use kilo code then you can just use client. But in this case we are going to use client and you can see this is the past conversation where I was trying to jailbreak this I model. So before you start a new task just go to your settings tab from here and then you can just choose code supernova from the model section. So you can find code supernova in here. And you can see this AI model has 200,000 context window and it does support images as well as an input.
So let's just go ahead and conduct our first task. So our first task to test the capabilities of this AI model is to create a 3D Rubik's cube simulation. So this is the detailed prompt and we're expecting to create an HTMLJS program using 3GS and it will use 3.js library to create a 3D simulation. So let's just see if it can create. So let me just press enter. And now let's just see if it can create 3D Rubik's cube simulation. By the way, clot sonet 3.7, clot 4 and 4.1 sonet, all of these models have successfully completed this task. So right now you can see it has already started writing code. The speed looks good and it has created this to-do list in here. The first thing that it is going to do, it is going to create HTML structure with 3.js, CDN and basic UI. The second one will be it will implement CSS styling for the application. then create main JavaScript file with 3.js setup. So you can see it is completing all of these task one by one in this to-do list. So let's just wait for it to finish. So far the performance is good but to see the final results. Let's just hold on and wait for it.
All right. So within less than couple of minutes it has just finished this task. So now let's just see let's just try to run and see if it can run this project and if we can have our 3D Rubik's cube simulation as expected. Okay. So now I have run this command start index.html. HTML and this is the index html file. You can see there is this script.js file as well. Then style dot CSS. So let's just see what we have. So this is what we have got. This looks cool. Okay, let's just try to scramble a key. Oh Jesus Christ. This is horrible. It looks good. Oh my goodness. It's it's just it's a mess. Let's just take a screenshot here and then ask it to fix this. You see test screenshot. So now let's just see what it does. Okay, so now apparently it has made some changings in code and it has implemented. It has fixed the cube positioning problem. Then there was canvas size issue. Then there was layer rotation logic as well. So these are the key improvements that it has made. Let's just try to run this project and then see what it has done so far. So all right, same issue. I think it is rotating from one origin point. So yeah and in fact if you just click on solve cube. Yeah, this is what it has done. So this is apparently the solve cube. If I just click on rotate up, it just goes like this. So it is rotating around origin this this particular region. Instead it it should have been rotating around this this origin. But yeah, so we can just like now further ask it to fix it. But I think it has terribly failed this task. I'm really having doubts about this model. I think this might be another Glock model.
But let's just see. Now let's just conduct more easier test because I just conducted the hardest test in the beginning. So now I have just given it this prompt to create an autoplay chess. So this is a detailed prompt to create an autoplay chess game. So this is going to be a simple 2D game. There's no 3D simulation involved. There is no use of 3GS library involved. So most of a models like Quen 3 all of those A models have easily completed this task. So let's just see if this code supernova can actually complete this task or not. So it should complete this task. So let's just wait for it and see how it performs. So apparently it has just completed this task. Now let's just run this script which is autoplay chess.html. So let me just run this script. And this is what it has created. Cool. That's that's amazing. Okay, it looks cool. Let's just see if it is making the moves. It's it's not making the moves. There is only one script. There is no other script. But this code should work easily. So this is comparatively simpler and easier than the previous task. Still it could not create autoplay chess scan. Okay. So it's not making any moves. It has just created like this front end but it is not making any moves. So far it has failed two task. I was not expecting it to fail the second task but this guy model has failed the second task as well which was comparatively easy. We can of course go back and ask you to like fix that issue but the main point was providing it a detailed prompt and then generating results in first attempt without any errors or any sort of hallucinations. So far this a model has just failed a real world coding task.
Now let's just try more simpler like maybe creating a simple landing page or like creating a dashboard design. I'm sure the same model will be able to do that because it can for sure create a landing page or create a simple dashboard but I'm sure as soon as we'll try to like use this for maybe backend development or maybe more complex task so it will fail again. So let's just conduct another testing and this time try to create more simpler project maybe a simple web scraping tool and then let's just see what this model is capable of. So I'm going to use firecrawl and then I'm going to use an API key from firecrawl. And now what I'll do I'll just create a simple web scraping tool by using firecrawl. So I'll just create a new folder. So I've just created an empty folder. So I've just got this API key from fire crawl. I'm just going to paste that API key. And now what I'll do I'll just give it a simple prompt. Please create a simple web scraping tool using firecrawl. We could use MCP server as well but let's just use simple API key that I'll provide a link to documentation. So let's just see if this a model can actually build a simple web scraping tool using fire crawl. This should be probably the easiest task for it. So let's just see just create a simple front end design where user can enter the website view and our tool should be to scrape data from that website. So it's a simple prompt please create a simple web scripping tool using file crawl. just create a simple front end design where user can enter the website URL and our tool should be able to scrape data from that website. So here is the API key firecrawl API key and let me just get link to the documentation of firecrawl. So if I just go to the docs I can just copy the link from here now it should perform this task easily without any any problem. This is literally it cannot go more simpler than this. I mean I can say okay even about the second task I know Kimik2 solved that second task easily. Gemini 2.5 flash has solved that task easily. Quen 3 has solved that task because I tested all of these different models on that simple task. So let's just see if if it can actually solve this particular task. It it cannot go more simpler than this. I mean if this AI model is marketed as the AI model for coding so it should work to create this kind of projects.
So apparently it has just created this tool. It has created an HTML front end with URL input from and CSS styling for the front end. Create JavaScript for front and back end communication. Create NodeJS backend server with express. Setup file cloud integration with API key. Create API endpoint for web scraping. Setup package.json and dependencies. So it should work fine. Now let's just run this command and run our project and see if it can actually actually run our project. So now it's running on local host. Let me just copy this URL to run this on local host. So, let me just open that. Okay, this looks cool. Let me just copy the URL of our website. Let me just paste that in here. Paste that in here. Scrape and cool. Okay, fine. Yeah, it has scraped almost all of the content from our website. Perfect. That's amazing. Perfect. It worked. There is no HTML content available. That's fine. Yeah. So, I can even copy it as markdown. Okay, cool. Perfect. Yeah, I mean it's working. It's working now. So finally this stealth AI model was able to create this web scraping tool because this task was way simpler than any of the previous tasks.
So after testing it I'm quite sure that this is not claude 4.5 set. This might be another model from Grock but I think for sure this is not claw 4.5 sonet. If you have any further questions just leave your comments in comment section. I see you next video.