OpenCode та Gemma 4 31b: миттєва та безкоштовна розробка повноцінних застосунків

Look at the quality of this build, guys. This is a 31 billion open-source Apache license model, and it's building at the same level as much better models. Now, I have a theory if you pair this with some other really good model like Opus or one of the better Gemini models, you will be able to build incredible websites pretty much for free. This model is currently free on Open Routter and I'm going to show you just how good it is in this video. Look at the quality of this. This is better than Opus. I'm not even kidding. The UIUX of this is better than Opus. The SEO here is better than Opus. Do you know how insane that is for a 31 billion model? This model has no business being this good. Let's jump into things. Now, this model has not been getting that much hype, and I'm not really sure why. Google Gemma 431 bill 250 262,000 context, completely free on open router as of today. I will say something off the bat. I can almost guarantee this model is better than GLM5.1, for example. It's better than Alibaba in my opinion, and it's completely free. Not only that, but this free version right here is actually pretty fast on open router uh on open code. I just built the entire website you saw before in 10 to 15 minutes with this model. Now, one thing I have to call out though from Google. I have to say I'm super disappointed by this. It's been almost a week since this model was released. I've been checking every single day to see if the model is available here inside anti-gravity and it's still not available. They have GPTOSs right here. They have Opus 4.6, Sonic 4.6, Gemini 3 Flash, Gemini 3.1 Pro. Where the hell is Gemma? Why is Gemma not inside here? Surely that makes sense for Google to have their cheapest model inside anti-gravity. They literally have GPT open source. Where the hell is Gemma? Sort it out, Google. Honestly, I don't understand what the hell they're playing at. With that being said, guys, let's jump into things. So, first of all, you want to make sure that you're on the latest version of Open Code. So, I'm just going to write open code. So, it's this command right here, open code models. And you want to write that. And then what you can do is you can actually look for Gemma like this. and just press enter a few times and make sure that you have open router Gemma 4 31B IT free. This model right here is the model that you need in order to use this entire coding system completely for free, guys. There's no cost here. Zero. Absolutely crazy stuff that this is even a thing. Not only that, but if you have a decent computer, you can run this on your local system and use it inside Olama. Now, unfortunately, my Mac is not powerful enough. The next Mac that I buy, I promise you guys, I'll get one that is powerful enough, right? But this one here is only 24 gig memory. So, unfortunately, I cannot run the new Gemma models on it. The next time I get a Mac, probably in about six months, I would guess, I will make sure that I get one that is powerful enough to run these models locally. This is a huge step forward for open source, guys. I'll tell you why. Because previously, these models were just nothing. They were complete trash. And only things like GLM and also Quen 3.6 were even close to being at I'm just going to say Opus level. It's cliche, I understand, but close to Opus kind of level. They're not Opus level, obviously, but they're close to it. They're as close as open-source models get. The problem with these models is just how large they are. This model here, for example, is 397 billion parameters, right? Which means you need 397 GB of memory, as far as I understand it. I'm not an expert on these things, but I'm pretty sure that's what it means, right? which is absolutely crazy. But as of today, you can run Gemma 4 31B if you just have a normal computer, right? So, a lot of people will have this kind of memory, right? Um, my computer upstairs, for example, has 128 GB of memory. I could easily run this model for free on Olama on that computer. right now. I don't because I'm using my Mac all the time, but just so you know, if you have a, I don't know, RTX 4090, probably even a 3060, 3090, something like that, you will most likely be able to run this model completely for free, right? So, all you do is you get OAM, you get the model, and then you connect Open Code and OAM together, and you can run this for free. Now, this is the prompt that I did. This is just a normal the school community prompt. I'll be doing more tests, more in-depth tests. I'll build a SAS with this. I'll build an Astro website with this. But this is always the first test I do. If a model cannot pass this test properly, in my opinion, there's no point to continue testing this model. However, it managed to smash it. Now, one thing I have to say is it did stop immediately and it said, "I can't even find my next app. Where is it?" This is a test to see how good models are at just continuing with things, right? because the prompt says you are inside a new next.js GS project, but there is no nextGS project. Right? So, this is a test to see how good a model is at dealing with this problem whether it'll just continue or if it'll ask for help. So, Gemma asked for help. It said, "Look, there is no what what the hell do you want to do?" So, I just said, "Start from scratch." And I said, "Don't use superpowers. I don't really want it to use um skills. I just wanted to do like a base test." So, I said, "Don't use superpowers. Just get on with it." And then bang, look at this. It just started building. Started building. Started building. Really, really nice build originally. really good technical build which is one of the main things I have to say and then it said I have successfully initialized and I didn't like this word initialized that means like it's halfass it basically right which is normal and then I said start the site and then there was a little problem where it was going to en which it fixed and I said needs a header and a footer and also just like 100 times more detailed then we're done thanks really simple prompt again bang header bang footer okay let's make some updates let's make it more detailed etc etc. Look at this. Bang, bang, bang. You could say again like keep making this more detailed, especially make the homepage. Wow, people, right? And then it's just going to go, it's going to continue. It's going to keep improving. So, I will say with this model, it's not very good at just running with it and building the entire thing from A to Zed completely from scratch without you prompting it a second time. But that's fine. I don't mind that if you have to prompt it two times, three times, it's free. Like, what do we care about bloody prompting it a couple of times, guys? Let's be honest. So, this is what I came up with. Like I showed you at the beginning of the video, the technical build is perfect. There's pretty much not a single dead link anywhere, which is really rare for a model. Not only that, it's done things like I really look at this. You never see little animations like this, right? If you gave this the ability to find images online or generate its own images with Nano Banana Pro, which is really easy to do. You just add it to a skill, this would look even better. So, imagine there are actual images here, which I'll do a test on that as well. Uh, let's just see if the Italian works properly. So, let's go here. Let's go here. Let's go here. Okay, Italian works perfectly. Oh, this is in English, though, actually, which is a bit of a problem. That's okay. Just all this means is that it didn't translate all of the content, right? It just translated some of the content. Let's see how the Italian is. So, no, alto luso permatrimonial in this place. Fine. Not bad at all. I would say this is minus points for not translating everything. I want to be completely open and transparent with people. Normally, I would expect this to have translated everything, but it didn't. This is interesting. This is a Claude thing. Claude does this all the time. So, you know, maybe maybe they train this on Wow, look at that. Maybe they train this on Claude's outputs similar to how other open source or how Chinese open source did. But look at this quality. Jesus. And the fact that it's like managed to put Salerno down here, which if you don't know, Salerno is like the beginning of the Costa Mapano, the Amalfi Coast in Campa, right? So, it's really interesting that it's got that kind of detail, that level of detail. It's got nice animations. Let's see how this looks on mobile. Let's see if it's completely messed up on mobile. It's actually not completely messed up on mobile. It's It's actually fine on mobile as well. Look at that. It's added this thing at the very top. Now this I will tell you right now this is a hallmark of Claude right when you when I see this I think vibecoded claude right so that's very interesting that it's put that my guess is Google has trained this model on the output of Sonnet 4.6 and Opus 4.6 six. That's just my opinion. It's not an an accusation, not legal advice, all that good stuff. But in my opinion, it looks like that they have actually just done the thing that the Chinese models did as well, which is train their models models on the output of Opus 4.6. I don't really care though cuz at the end of the day, this model is small enough to run on a bloody computer, a normal computer, probably your computer if you go and look right now at how much memory you've got. and it builds as well as Opus. The technical build here is perfect. I need to see how this builds a SAS and whether it's possible, but I believe it would be. Just before we end the video, guys, just a quick shout out from our sponsor, me. This is harbor seo.ai. It's a tool that I've been working on for about 2 years now, but in the last 3 months, we've shipped more updates than were shipped in the first 2 years of this product. For example, you can now scan your website for site health and technical SEO issues and then fix them automatically with Harbor AI, which will literally inject metadata into your website and one-click fix your entire website's metadata. Not only that, we recently updated the writer to increase the quality of the writer. Now, it uses Sonic 4.6. And the pricing, guys, it's Oh my god, look at that. 431 clicks. What the hell? Crazy. That's gone up like 150 in just a few days. So, 107 pages published, 37,000 impressions, 431 clicks. This is getting real clicks for real businesses and helping people make sales every single day. Now, the pricing is 29 a month, 25 articles a month. It's more than fair. Unlimited articles, unlimited research runs, 49 a month, 50 articles a month, unlimited sites, or 99 a month, 100 articles a month. You will not get better than this product right here. I'm telling you, this is good for a SAS project. This is good for any website, any business. It can scrape anything and write contextaware content for your business. It doesn't matter what CMS you're on. It doesn't matter where you're hosted. This thing will write amazing content for you. And we have huge ideas coming such such as agency which is coming very very soon which will basically allow you to run an entire agency in a box, write content for people, get back links, create pages for people and also content clusters, topical authority, spy on their competitors and much more for $29 a month or €29 a month. There's a link to Harbor in the description and in the pin comment. Thank you so much for watching, guys. If you are watching all the way to the end of this video, you're an absolute legend and I will see you very, very soon with some more content. Peace out.