Alibaba's New $10 Coding Plan is ABSOLUTELY INSANE (MiniMax 2.7)

Hello and welcome to this video where I'm going to be talking about the KOD coding plan. But just to quickly go through this guys, there was originally this Alibaba offer of $10 a month, right, for really really good usage. However, this is no longer available and the minimum you can now pay is 50 bucks as far as I understand it. Feel free to correct me if I'm wrong, but that is what I have heard. However, I wanted to kind of talk to people about this because you can get a two week pro trial with 300 credits and for 10 bucks a month, you get 2,000 credits on Kyoda. And Kyod for me has actually been extremely powerful. So, you can choose from all of these models here. Miniax 2.7 uses 0.2x models, right? So, if you want to use some of the cheap Chinese coding models, I would say that Kyoda is actually the number one choice for cheaper coding. So, in this video guys, I'm going to show you how to set up an entire project and we're going to test whether Kyod actually has the goods or whether it's not good enough at the moment to be used as a coder. Whether or not this works doesn't matter because at the end of the video I'll tell you whether I think it's worth using this for 10 bucks a month or if you should stick to something like free codeex or anti-gravity which they are both free but they probably won't be free for very long. Now just before we jump into the video guys quick shout out from today's sponsor me. There is currently a timer on harbor seo.ai. It's almost run out guys. I I I might end up extending it again. I don't know yet. I just want as many people to get this founder offer as possible. I don't want to I don't want people to feel like they've lost out, right? So, I don't know if we're going to be extending it, but now that we've actually got kind of what founders are going to get, I kind of worked it out last night through a series of quite a lot of working things out. Basically, you're going to get more articles per month. You're going to get six times rollover. So, um if you don't use all of your credits for 6 months, they'll roll over six times. And also, the price is locked forever. We also added a new plan which is the growth plan which is $49 a month or sorry $49 a month and that gives you 50 articles a month, unlimited sites, unlimited research bait and link magnets included and monthly limit resets each cycle. Although that's not actually true. If you sign up as a founder, you get 6 months of roll over tokens. So if you don't use your tokens immediately, doesn't matter. We also, as I keep saying, have just increased the quality of the writing. The writer is significantly better now. So this is an example of one of the articles that Harbor writes. It's just super high quality now. It chooses the images better. There's no more duplicate image issues. It does these fantastic tables which will rank on LLMs. And yeah, guys, go and check it out now. There'll be a link in the description and in the pin comment of this video. Go and grab Harbor while you can still get the founder pricing. Thanks for the attention. Let's jump back into the video. Okay, so Kyod, guys, this is Kyod. I actually kind of do like this. I'm not going to lie. So, you can set up agents and skills here. I'm probably not going to do that in today's video because I just want to see the raw coding power of a tool like this. I'm going to give it the best chance I can possibly give it though. So, let's just jump into things. So, I'm going to show you how to do absolutely everything. So, the first thing I do is I go on Google and I type in convex next.js. And then we grab this next.js quick start right here and just press copy. And then we'll paste that in. And then we'll just call this uh my app. That's fine. We'll do Nex.js. We'll do clerk. And then while that's setting up, let's just go to clerk. I do this all the time, guys. I've shown this process many, many times. But just follow my process, guys. You'll be able to build an app with literally anything. This is Harbor, by the way. This is what Harbor's looking like right now. Super, super exciting times for me. Almost 3,000 users now. Um, but yeah. Anyway, create an application. got this Kyod test v2 because I've probably already got Kyoda test somewhere. The good thing about Clerk is it just lets you jump in. It's really really nice. You don't have to worry too much about things, right? Like you can literally just test it locally very very easily. Okay, so we've got the API keys here. We're not actually going to give them the API keys this time. What I'm going to do instead is I don't want to do too many things at once. So let's just quickly cd my app. We need to start the convex application for the first time which builds the convex project. Right? So we'll call this coder test v2. Right? So now this is creating the convex instance if you don't know. Oh uh yeah. Okay. Oh, nice. Setup convex AI files. That's new. Really, really nice actually because now it actually has skills right here. That's actually super super nice. Okay. I might delete those though. I'm not sure cuz I wanted to use it. I I wanted it to be able to use it itself. It also looks like it's in the wrong folder. Uh so we would probably have to move those for it to use, but it's kind of annoying that it did that to be honest with you because that kind of spams my my stuff, but whatever. So now that we've run that, what that does is it adds the convex deployment to the M file. So all we have to do is just add our um variables here. And then there's one more thing we have to do for this to work. So we're going to go to configure here. We're going to go to JWT and add new template. I don't know why they didn't just do this out the box honestly. Um there should be an easier way to do this in my opinion, but we copy this and then I don't actually know what the end point is. I don't know what the variable is actually called. So let's just check the documentation quickly. Yeah. So I'm on um I just Googled convex and clerk and then search form. So it's this here clerk JWT issuer domain equals. And then we need to give this here. And then we just save. That's actually everything set up. Like it can literally build you anything now from this if it's a half decent model. So let's see how good this model is. Um I don't know whether to use agent or expert. Automatically assemble an AI engineering expert team for each request with multiple expert collab experts collaborating in parallel for efficient high quality results. I don't know if I'm going to use that right now. I definitely want to test that. That is very very interesting to me. I think what I'm going to do is I can't choose which model to use. I'm just not sure. So maybe I'll go auto smartly select the optimal model balancing performance and cost. It's probably just going to use Quen every time, right? Uh let's just use Miniax, I guess. Let's see how Miniaax actually performs here. So let's say um please set up clerk fully for me. I have set all the variables but I want them to be redirected to slash dashboard when they an account and also you need to do the convex schema to make an account slash user when they make an account. So that should be enough for it to basically build everything that I need. Now, there's a couple of ways that you can go for the kind of test. Oh, sorry, the the design of things. So, I like v 0ero personally because it uses shaden, but you can also use stitch as well. And just another thing, this is convex. Convex is absolutely amazing. It's similar to superbase if you think of it like that. So, we have the data here and the functions. The reason I use convex is because it's like a hosted backend. So instead of being on my server, which is probably quite easy to hack or, you know, quite vulnerable to attackers, instead I'm using Convex, which should theoretically just be safer out the box just because, you know, they have a vested interest to keep my stuff safe. Now, that's not advice or legal advice or anything like that, obviously, but I'm just saying probably just by odds, their system is going to be more secure than me just running my own NextJS server, right? So, just so you guys know, normally when I build something with this stack, I use the agents SDK from OpenAI. The reason I use this instead of the Claude SDK is number one, cost. Um, the cheap models from OpenAI are better than Haiku 4.5 by a long long long way. And the other reason is it can run in a serverless or like hosted backend environment, whereas the Claude SDK actually can't run in convex, as far as I understand it. Maybe I'm just doing everything wrong, but as far as I understand it, that is true. So, they have a TypeScript thing here. I I can't find a skill that will just do this for me, which is kind of annoying. Um, but I'm going to be making a fairly simple AI application. The idea here is not to show you how to make an AI application, but it's just to make an entire project as quickly as possible with a very, very cheap model. You guys have your own ideas, so it's not going to be like a crazy idea or anything like that. So, we will use Stitch because Stitch, as far as I understand, is completely free to use and they did recently have an update which I thought was pretty good to be honest. So, we're just going to 3.1 Pro and then we're going to go. Okay, so there's one thing, one more thing we have to set up locally which is um you'll see this error quite a lot. Basically, I set the M variable here in the wrong place. I actually have to set it inside convex. There are two places to set variables. You're probably familiar with this concept using Versel dishloian etc. Um, sometimes you have to set it in superbase and sometimes you have to set it in your m.local file which will then become the versel file in the future as well. So this should already be set up with just basic clerk authentication. We're about 3 minutes in here. So let's just see. Okay guys, so already you can see that the slash dashboard is loading. This is honestly just as good as something like Opus at getting to this point, which is pretty interesting to be honest with you. So, I'm just going to use Stitch here to just do a very quick ideation/design. Like I said before, you can also use Vzero. Like I might even just copy this prompt and just whack it into VZ at the same time. You could also use Google AI Studio as well. Like I'm just giving you different options um that you, you know, people might be interested in. Okay guys, so we've got a couple of designs here. I'm not really the design guy. If you want designs, maybe go to a different YouTuber. Uh, to be honest with you, there are plenty of good design um, YouTubers on this in in this space. I'm just going to probably go with the Shad Cien one here. Um, let's just see if it also did the dashboard. I do prefer Vzero to be honest with you to Stitch and Google AI Studio is is still building. So, this is really, really nice, right? Obviously, it does look very AI generated, but at the end of the day, like I'm just doing an example. So, and it looks better than most AI generated things. So, let's go with this one here. Now, let me just see. I should be able to press some kind of download here. Okay. So, you click on the code thing here and then you just press download zip. That should download it as a zip. Beautiful. And then double click. Okay. And then go to Kyod. What we want to do is we want to drag this into the Kyod space like this. There we go. Add folder to workspace. Trust it. Yes. Allow. Yes. Okay. So, now we want to at@b ytf. It's not coming up. Oh, there it is. Okay. So, you can normally drag and drop. Let's see if we can drag and drop this into here. There we go. Beautiful. Use this file to add the front end plus um dashboard. Uh use the shad CN design, etc., and implement it on top of clerk, I should say. Okay, so Miniax has completely decided to ignore everything I gave it before. I mean, that's definitely a markdown in my opinion. This is this is just not what was given to it at all. It says it's done the sign-in page. Yeah, that's still not the stuff that it was given at all. Beautiful dashboard with Yeah, sure. Okay. really beautiful. Uh, it just hasn't really followed that prompt. That's okay, though. I mean, this is a test at the end of the day. So, let's see how it does with the AI stuff. If you're curious, this is what AI Studio came up with at at the end here. So, all you do here, I always go to AI Studio and then go to the playground and select I would use Gemini 3 Flash just because it's quick and easy. Press get code here and then go to TypeScript and just press copy and then say um please implement the blog post writer. The user should give a keyword and it should just write a blog. Again, this isn't to show you guys how to make like a specific AI tool. So, this is not going to be the best blog post writer in the world. Obviously, Harbor SEO is the best blog post writer in the in the world. But just to show you guys how this would work, right? Again, so we also need an API key. So let's just go back to um go back here, press back, go on my page, then once I'm here, get API key, and then just press copy. I'm not going to show this guys cuz yeah, I can't show this. But you can see here it's put it here already. So I'll just put this here. So one important test to do guys is to see whether or not this actually created a user. I've done this test so many times now and one of the most common mistakes that one of these AI models makes is that there's no user table. So there is a user table and you can see that there is actually a user here as well. So it has followed the instructions pretty well. Honestly like for what I'm trying to do this is like opus level. I just showed my API key. It's okay. I'll delete this one and start a new one. It's fine. But when I say opus level I just mean at doing this task right which is a step by step. I'm watching everything. I'm doing everything myself. Step-by-step task, right? It's not, you know, adding a huge feature to Harbor SEO. It's just literally just a really, really simple task. So, when I say it's Opus 4.6 level, that's what I mean, right? If you do the same test with Opus, it will give you pretty much the same result that I'm getting here. The only difference being um the it would have followed the Shaden design much better, I would say. Honestly, this is pretty slow as well. Like, I don't know if I can complain about that. without sounding like an idiot, but compared to claw code, this is hella slow. Like really, really slow. Okay, so again, I mean, it's handled it pretty well from what I can see. This looks like it's actually making an API call. Again, completely not followed the Shad CN prompt whatsoever, but so and obviously the content is going to be absolutely terrible. Okay, it's just an example. Overall guys, I mean, if you want a budget way to code in 2026, I think Kyoda might be one of the better options. There's still a few things I need to test. Like, I'd quite like to test just using it on auto. I'd quite like to test using quest mode and I would also like to test using experts as well. And also, this looks really good. Upgrade now. Upgrade to pro plus ultra teams. I might bite the bullet here, guys, and grab this at some point and do some more tests. But like overall, honestly, this was as fast as getting this all set up as Opus 4.6. Now, the only thing I would say is obviously the Shaden hasn't been followed, and if I was going to do like a complicated AI implementation like Harbor, I would definitely rather use Claude code and Opus 4.6. But what I'm trying to get across is if you do want a cheap way to code, this might be the way forward. Thank you so much for watching, guys. If you are watching all the way to the end of the video, you're an absolute legend. I'll leave the video there. Check out harborseo.ai and I'll see you very soon with some more content.