Transcription
Hello everyone, and today I will tell you how to use code with the JLM 4.5 model much cheaper than with Tropic plans. Also, there is a subscription, and all of this is easily integrated into CLD-code. That is, we will simply replace the model with the JLM model and save a lot of money on this. Yes, I will warn you in advance that I am currently sick, I have a runny nose and a sore throat, but I hope this does not significantly affect the perception of my voice. In general, let's begin. So, we all know what CLD-code is. That it has proven itself as a powerful agent for programming in the terminal. But it has a problem in that the model's performance is unpredictable. What Tropic recently admitted is that random errors sometimes occur and the quality of responses can decrease at different times of the day, for example, due to an influx of people. Well, and strict, expensive limits on usage. That is, if we look at the tariff plans that we have, where, accordingly, there is our code, free code is not available, it is only with subscriptions. Here, $17 is for paying for the whole year at once. Otherwise, it says $20 for monthly payment. And with this, we get plus or minus 45 requests in 5 hours in CLD-code. This figure is not exact. Tropic, as far as I remember, does not share it themselves, but approximately this is the amount for 5 hours. Again, all limits are reset on the Max tariffs. On the Max X5, it's up to 200 requests. Well, and prices there start from $100, respectively, these are inaccessible amounts for most people. And Max X20 costs $200, as I remember, and more. And there, in principle, there are not many requests either. After this, JLM comes to us, who have finally decided to create their own subscriptions. Not just access to models via API. Even if we take the API, JLM 4.5 models are maximally cheap. They decided to create coding plans, which are also a monthly subscription, a fixed cost, and we also get the same system as Entropic, that is, we get a certain number of requests and a reset of limits every 5 hours. Let me show you right away. We just need to go to their website z.Ii. Their domain is maximally simple, it will redirect us to their chat. Here you can try the models. JLM 4.5 is available here, their top model at the moment, the Air version, which, well, roughly speaking, is a flash, a mini-version of the same Geni or GPT, just cheaper, faster. Well, and other models are also available here. You can try them out. We are interested in this tab API. We click on it, and it redirects us to their documentation. And here we can scroll down to see the list of their available models, there, agents, and so on. But here at the very beginning, we are immediately interested in this GLM coding plan panel. They literally just added this the other day. We just click on view plans. And what do we see? First of all, they have discounts for the first month. So, for us, it's only $3.15 for the first month. GLM 45 integration with CLD-code is immediately available, maximally fast. Well, and here you can look at quick answers to questions. In general, what's the point? If we take their tariffs, then when paying for the first month, we have only $3 for their Lite tariff and $15 for the Pro tariff. Further, when paying for the second month and so on, it will be $6 and $30. But still, $30 is slightly cheaper, or rather, slightly more expensive than CLD Pro. The cheapest tariff from Entropic. And at the same time, if we remember, CLD Pro has plus or minus 45 requests for $20. For $30, we will get up to 600 prompt requests, to be precise, in 5 hours, which is simply many times more than OPIC provides us. If we look at the first month only, it even turns out to be cheaper. For $3, we get 120 requests. Accordingly, this is still almost three times more than Entropic at a price more than six times less. Well, this is again only for the first month, but still. Well, also, perhaps you will immediately have a question, how to pay for all this from Russia. Now I will show you one service that I will use this time to pay for subscriptions. I will also show you how to pay for a subscription to CLD-code, just in case, if you are still interested after this video. And we will return to JLM. I will now use this website. When you simply follow the link to the main page, you just need to click "Issue Card" and select "for purchases and subscriptions." Also, if you use my promo code, you will get a 50% discount on opening the card itself. Then you simply select the BIN here, you can leave any here. You select the top-up amount in dollars and the payment method, accordingly, you also choose. Then you place an order, and you will get a card. Here is my card. I will log into it. Here all the data is available. I simply return to CLD. I select "Get Pro Plan." I select monthly. I already have a card here, I will remove it. And now I just need to fill it in. United Kingdom is selected here, address, I just copy the card number, everything is filled in here too, subscribe and pay, and everything is fully subscribed to pro. Now everything will update here. I have the pro plan. I just paid for it. My card, and that's it, I can continue to use my subscription with CLD-code. So, after all the manipulations with CLD, I have $43 left on my balance. This should be enough for me. So, I just return, click on subscribe, enter the card number and the rest. In general, I will fill everything in now and return. So, everything, I have paid. It gave me "Subscribe successful" and that the subscription will be auto-renewed on the 13th of next month and $6 will be debited, which I don't have on the card, of course, but that doesn't particularly bother me. Accordingly, here, with CLD-code, everything is done. After this, I go to API management in API keys and create a new API key. Let the name be test. Confirm, and that's it. Just copy it. I return to the terminal. I will close CLD-code for now. So, what we need to do now is to change the environment variables, the base URL of the API, and, accordingly, add our key from Zi. That is, we change the endpoint to Zi and, accordingly, we also change our key. So, we write export. I have MacOS, on Linux it's the same command. On Windows, you just need the set X command instead of export. Everything else is the same. If anything, I will leave the full instructions in Telegram, as always. So, base URL equals and https. Seems correct. /ip/entropic enter. Next, we need to change the token. Our authentication token, or in other words. So, we write like this: token equals and paste the API key that we copied from the site. I accidentally pressed minus instead of equals. And after that, we also write our CLD command to launch CLD-code. And then, to check that everything worked, we write status. And now we have the account via token and API configuration, Entropic base URL, and our API, accordingly. I also forgot to mention that we could have gone here to developer documentation and examples. After which, we have here the options for where and how to integrate our GLM. That is, here we could have done it all with one command like this for CLD-code, which we would simply paste into the terminal. We used this as an alternative. Well, and here it is also demonstrated how we can quickly change models. That is, by default, CLD-code uses two models. This is Antropic Model as the main model and Entropic Small Fast Model as an additional model for any scenarios, like file search syntax checking. By default, JLM 4.5 and JLM 4.5 Air are used. We can, here in this directory, create a configuration where we change the Tropic model, for example, the main one to Air or vice versa, the Small Fast Model, we can change it to JLM 4.5. For example, my terminal currently says that my default model is CLD S 4. What I will do now is try, first of all, to restart the terminal. Then I will re-export the variables and also change the opic model. That is, essentially, just like this, base model. I will double-check. Here, they are simply not here, model. Entropic model equals JM 45. Essentially, it should be like this, right? JM 45. Export. Again, CLD. Okay, now it worked. Now I have the GLM 4.5 model. Accordingly, if you don't want to type all this, again, you just need to go, as it says here, to the documentation, to this directory. That is, this is your user, in my case it's forget me. And create a settings.json file here and write this into it. So, paste it here and that's it. You can also add here what we exported at the beginning. That is, our base URL and token. And you won't have to re-enter all this every time in the terminal. That is, you just put a comma here, write base URL and so on, and change everything, accordingly, to your data from JLM API. Well, and so that I don't just show all this, I will now ask one and the same question to two models. This is, accordingly, CLD 4 SAT and JLM 45. I will throw the same request into both CLD-codes. And we will compare the results. So, I created two folders. The first one is JLM test. The status will simply be the model, accordingly, JLM 4.5. Now I will write the prompt on which we will compare everything. I wrote such a prompt. I will also copy it for myself so that I write the exact same prompt for Cloud 4 Sound. I just press Enter, and that's it, it starts working. Here I created a separate folder for testing CLD-code separately. Accordingly, here I write status, and here is my CLD Pro account. Accordingly, this is the CLD Son 4 model. And I paste the exact same prompt here. Now it will also do it for me, after which we will compare the results of both models. Directly head-to-head. What's interesting, JLM 4.5 has already finished on the left, as you can see. Open index.html in the browser. Of course, it's probably bad. I thought of it this way, I made two windows. It's not clear here, but now let's open index.html. So, this is the JLM test folder. Here is html. I open it. So, Image Editor Pro. So, drag the image here or click to select a file. This is again from JLM. So, I upload the image and nothing happens. If you drag and drop, oh, it works by dragging and dropping. So, here is the image I uploaded. And here we have options to crop, reset filters. There are some sepia, blur. True, they are not applied, as I understand it. Let's say I apply a blur. I don't need a watermark, if I crop. And like this, I select an area. It's a bit inconvenient for me, though. Yes, it's strange. It's like I'll select it somehow like this. I can crop it crookedly. Well, if I rotate and add a watermark here. For some reason, it decided to add the watermark not to the application, but to the exported image, but that's fine with me. I'll export it. And yes, it turned out exactly as we had in the preview, but none of the effects were applied. That is, the brightness and contrast I selected here, it didn't change anything. Well, I don't know, maybe because the image is black and white, but in any case, the reset button works, and the cropping is crooked. But it all looks quite good here. It did it very quickly. I was just launching CLD-code the second time, and it had already done everything here, everything is great. Meanwhile, our Sonet has already finished. Let's go to it then. So, in CLD-code test. Here, there are no problems with just uploading images. There, with JLM, there was a problem that the image was only uploaded by dragging and dropping. And it added a watermark to the page itself, by the way. Here, you can also see the watermark in the page title. And what can we do? Accordingly, here we have crop, which works like this. Convenient. Apply crop, and everything worked. Rotate, flip, and also settings. It added all these brightness, contrast settings. Well, if I apply sepia, vintage, look at the preview. None of these effects work. R, all of that works. Accordingly, I can also click rotate. For example, I'll take vintage, crop it somehow like this. Apply crop and export. And none of the effects, as I understand it, were applied here. I used vintage, but there is no such vintage here. And it also decided to put a watermark on the image. Although I didn't mean that. Well, okay, never mind. But, by the way, I like the interface here less. For me, JLM's interface turned out to be cooler somehow. I don't know, this is just my personal opinion, of course, but the functionality works better with CLD 4, because here we have problems with cropping. These effects also don't work, but in principle, they don't work here either, it's not a drawback. But again, personally, I like the interface more here. Here it's somehow, well, so-so, personally I don't like it. But here it added additional Rotate and Flip functions, which also work. JLM doesn't have that. In general, I will finish with these two models. What can be said now? I haven't tested JLM 4.5 very thoroughly yet. I can't give any specific answers. What I can definitely say is that JLM 4.5 has a smaller context than Claude, 128,000 versus 200,000, which will affect it. But what is also important is that JLM has a very fast speed. Really very cool. It writes quickly. In general, yes, it's worse than CLS 4. It coped with the example that I ran just now. But if you use them together, in my opinion, it will be an incredible power, because, let's say, we lay out the framework, the interface, the main features, just like placeholders, some functional capabilities in the interface using JLM, because it does it quickly. And then we finish the functionality with CLD-code, because it does it qualitatively. In my opinion, such a combination is an excellent option. And at the same time, we will have better limits. That is, let's say, we can combine a $3 subscription and a $20 subscription, and we will have, say, 120 + 45, that is, 165 requests in total in 5 hours. And this is a saving of plus or minus $80, if we compare it with CLD Max X5. Well, and you can generally check JLM further if you don't want to spend a lot of money on CLD Pro, that is, $20 or $100, it doesn't matter, because the maximally cheap $3 for the first month, well, that's just pennies, literally. At the same time, we get 120 requests in 5 hours for a month. This is incredible, in my opinion. But again, this offer is currently limited, so perhaps when you are watching this video, it will be $6, but it is still maximally profitable, more than three times cheaper than the cheapest subscription from Claude. In general, it's worth considering for application, it costs $100, because it's very cheap, very fast, and quite high-quality compared to the price. I will finish here. Subscribe to the Telegram channel, because there is a lot of useful content there. Proniro city. I hope you found it interesting and useful. Thank you for watching.