Transcription
Hello everyone. My name is Kostya, and quite recently Alibab Group released a new version of Quen for us, namely Quen 3 CodР Next. This is an open-source model, meaning each of you can download it to your computer and connect it to your code editor using Ollama. Or, like me, you can use it through Open Router. It costs mere pennies here, only 7 cents per 1 million input tokens and 30 cents per 1 million output tokens. This is extremely, extremely cheap. If, by chance, you don't know what Open Router is, it's a specialized service where you can use many neural networks. There's a whole bunch of them here. We can see which ones are available. So, there are very, very, very many of them. I'll get tired of scrolling. And you can use them for your applications, for example, for OpenClow, for Kilocode, for clklokod. You can also top up all this with crypto. I think for some countries where there are card restrictions, this is quite a profitable option. Another interesting point is that once you top up your account here with 5 dollars, you can use many free neural networks and access them. I believe you can make 50 requests per day if you've topped up once with 5 dollars. Ah, maybe it has already increased. I checked this information a very long time ago, but free access is available right now. I will test our neural network in Kilocode. For this, we need an API key. I will attach the link for you. We create a new API key. Let's call it, for example, test. I create it, I copy this value. Then I go to antigravity. You can use Visual Studio Code, Fry, and any other editor. There, you can type Extension, for example, Kill Code. And here you will have Kilocode, you just install it, and you will have this editor. Here you will need to add a new profile. I already have one added. You will add a new one. Sign it somehow. And here, in providers, you need to find Open Router. I already have it here. You will paste the key here. I will paste it now. And among the neural networks, you will need to find 3 code Next, somewhere here, because there will be many quendor here, and specifically, code Next is the only one here. And we will experiment with it today. So, to begin with, I will use something, well, very simple, like create some kind of landing page. I have prepared this prompt, like create a landing page for an IT company in the service sector. Create, well, the creation of mobile applications, come up with the blocks yourself. The design should be modern. These themes in the background want 3D effects. In general, I wrote that it should use HTML, CSS, and JavaScript. So, nothing grand like React is needed. So, I send all this, and now we will have thinking and so on. It will start creating something now. As soon as it creates it, by the way, it tells us how much money we have used, if anything, and as soon as it creates it, we will look together at what happened. It was very fast, it probably worked for only a minute, and it's already done. Here is this project. We will now open Chrome, and I will just drag the HTML into Chrome and see what we have. For a neural network for pennies, this is, I would say, a super result. Believe me, I work with neural networks and I know what kind of designs they make. This is, this is a super result, one might say. Well, yes, like. And it works, yes? It's even, hmm, not bad, not bad, not bad, not bad. So, what? Let's look at the flaws. No, actually, even here everything is fine. It even made some kind of 3D background. Well, there are a few glitches with the text, but that's nothing. It's probably not adapted to work in Russian. What it has already produced is already not bad. Well, landing pages are landing pages. That's the easiest thing neural networks can do. Let's do something more complicated. Let's suggest it make a web application. So, I created a new project. I will open it now. Open folder. And here is my new project, completely empty. I will open Kilocode now, and we will try to do something again. So, I will turn on everything on the orchestrator. I have it on in all modes, if anything. And we need a new prompt. But first, what's best to start with, any web application creation, we will tell it: "Create an empty Next JS project." Because during the process, some bad things might happen when it creates the project with a large prompt. Therefore, it's better to tell it in advance, let it let it create the project first, and then we will enter the prompt. So, I wrote this prompt. Create a calorie counting application. I will enter products, pre-installed in a local database, and select grams. The web application should calculate their BJU and save it in the local database. The design should be, ah, as modern as possible, using emoji, icons. In general, bang, send. And now let's see what we get. As soon as the neural network finishes, we will again look at what it will do for us and whether it will create a web application at all, because the neural network is very, very weak, but but so far it's showing itself very well, especially you, like, understand the fact that you can install it locally and pay nothing at all. You just need suitable hardware, of course. By the way, I also wanted to discuss one point. Why install local LLMs on your hardware at all, if you can use them by subscription or in some other way? In some companies, I have an acquaintance who works at a factory, and she is forbidden to use any cloud-based neural networks. That is, she can only use neural networks installed through Ollama and connect them to her editor and only use the neural network locally to help her write code. So, I see that the neural network has finished its work. It asks us to go to localhost:3000, and we will do that now. This is what we got. Let's try to select a product. Well, okay, kefir, for example. Aha. We have small problems with fonts, or rather, with their color. Well, I drank 500g of kefir. So, kefir has been added. Okay. It calculated the calories. Good. Is there anything more serious? Pasta. Well, let's eat pasta. 300 grams. Uh-huh. Calories, everything works, everything is calculated. Buckwheat. Well, let's eat a kilogram of buckwheat. Not bad, not bad. Let's delete the buckwheat. Aha. Well, this application works. It needs, it needs to be given credit. I just reloaded the page. The data remained. Uh-huh. So, it saves some kind of local database. So, what do we have here? Mo, well, we have a small number of products here, their quantity is limited. Tomato, I don't know, ate 100g, calories, everything is growing, everything is being added, everything is being removed. In general, it coped with this. Even, well, the design is quite normal, of course, not the top one, but it's quite suitable. There are small micro-glitches, but that's nothing serious. For its capabilities, I remind you, it only has 262,000 context, and it costs, well, just some minuscule amount of money, considering that you can also install it locally. Well, it's like a super solution for some people who need to use local LLMs. And we have only checked it for creating some websites, yes, and small web applications. But if you put it in open glow, it makes sense. I will probably install it, try it, because, well, it consumes very little, but it does some pretty good things. So, let's summarize. What do we have? We have a decent LLM for very little money, or even free if you install it locally, which can be used, which shows, well, quite good results. Be sure to write your opinion in the comments about it. Maybe you have tried it or will try it. Give it likes, in general, write comments, subscribe to the Telegram channel. See you soon. Bye.