OpenCode + DeepSeek Flash V4 = Full App INSTANTLY (FREE*)
Chapters23
Introduces DeepSeek V4 Flash pricing at $0.14 per million input tokens and $0.28 per million output tokens, noting its affordability and popularity despite requiring advanced hardware or routing to run.
OpenCode + DeepSeek V4 Flash delivers a surprisingly fast, cost-effective full-stack prototype, showing off a full frontend and backend build in minutes using open-source flash models.
Summary
Income stream surfers digs into DeepSeek V4 Flash's promise with OpenCode, highlighting how it runs a complete orth project through Superbase and OpenRouter for a sleek frontend and backend in minutes. The creator emphasizes the model’s price point—0.14 per million input tokens and 0.28 per million output tokens—arguing it’s among the most affordable options on the market. He demonstrates real-world speed, showing 20 cents spent for substantial token usage and 11% of the context used, while assembling a live dashboard with a faux sign-in flow. DeepSeek V4 Flash is praised for high token throughput, impressive speed, and competitive performance relative to Gemini 3 Flash, especially given its 1-million context window. The discussion touches on model efficiency, openness, and the potential for open-source alternatives to challenge monopolies in AI. The demo extends to frontend design and backend integration, including creating an Astro-based orth project with sign-in/out and a Superbase-backed dashboard. The video closes by praising the open-source route and recommending DeepSeek V4 Flash for developers experimenting with flash models, while also plugging Harbor SEO.ai as a separate tool. The creator anticipates producing more content focused on this model and its implications for open-source AI tooling.
Key Takeaways
- DeepSeek V4 Flash is cheaper to run than Gemini 3 Flash, with significantly lower input/output token costs.
- The model demonstrates fast token throughputs and can power full-stack apps (frontend + backend) using OpenRouter and Superbase in minutes.
- Open-source flash models can deliver practical, end-to-end demos (sign-in, sign-out, dashboard) with minimal setup and clear cost benefits.
- OpenCode and OpenRouter enable rapid prototyping of both frontend and backend components around a flash model.
- Despite occasional hiccups (e.g., sign-in flows or server stability), the workflow shows how quickly a production-like app can be assembled.
- The creator advocates for open-source AI tooling to counter monopolies and expects more open models to emerge, potentially reshaping the market.
- Harbor SEO.ai is promoted as a cost-effective SEO tool, illustrating how the video blends AI tooling with practical business applications.
Who Is This For?
Developers and AI enthusiasts who want to prototype full-stack apps quickly using affordable, open-source flash models. Great for those exploring OpenRouter, Superbase backends, and fast frontend scaffolding.
Notable Quotes
"This is one of the most reasonably priced models on the market. So, this is obviously a very popular model."
—The speaker highlights affordability and popularity of DeepSeek V4 Flash.
"For it to be number 17 in programming already, very, very interesting."
—Noting the model’s growing adoption and ranking in a specific category.
"This is fully connected to my Superbase, right? So that's how it's actually, you know, making these changes."
—Demonstrates end-to-end integration with Superbase for live changes.
"It's stupid cheap. It's stupid fast."
—Emphasizes the core value proposition of the Flash model.
"I think this is my new favorite model, guys, to play around with."
—Personal endorsement of DeepSeek V4 Flash as a go-to tool.
Questions This Video Answers
- how does DeepSeek V4 Flash compare to Gemini 3 Flash in price and performance?
- can you build a full-stack app using OpenRouter and Superbase in minutes with DeepSeek V4 Flash?
- what makes open-source flash models faster or more efficient than traditional LLMs?
- how can I run OpenCode projects with DeepSeek V4 Flash on open-source hardware?
- is Harbor SEO.ai a good companion tool for developers using AI to create content and apps?
DeepSeek V4 FlashOpenCodeOpenRouterSuperbaseAstroClaw Code RouterHermesOpen-source AIGemini 3 Flash comparisonHarbor SEO.ai
Full Transcript
Do you guys understand how ridiculous this model actually is? Deepseek V4 Flash. Look at the pricing of this model, by the way. It's $0.14. So, 14 cents per million input tokens and 28 per million output tokens. This is one of the most reasonably priced models on the market. So, this is obviously a very popular model. This has 281,000 downloads this month alone, which is absolutely crazy. Now, obviously, it's a big model, so you're not going to be able to run it on kind of standard hardware, but you can run it through open router. And today, I'm going to show you just how effective this model actually is for its price point.
And it's pretty damn good. So, I just want to show you something, guys. This is open code running this new flash model. First of all, look how fast it is. I've spent 20 cents right now. I've used up this amount of the tokens and that's 11% of the total context. But this is real. This is a real backend coding example which nobody does by the way on YouTube. So basically what I've done is I've created a full orth. It's not running so just give me a sec. I've created a full orth here uh using um Superbase, right?
So, it's just doing some changes on on the on the set on the colors and things. Right now, it seems like there's no sign out. I know there is. So, you can see get started up here. So, that's I'm just going to click that. So, we'll just do a kind of fake email just to show you guys here with a password. Okay. So, I need to sign up first. There we go. And then I get automatically logged into the dashboard. This for 20 cents is crazy. Okay, it's really really interesting stuff, guys. This is fully connected to my Superbase, right?
So that's how it's actually, you know, making these changes. I'm just trying it with some front-end stuff right now, but just overall this seems like a very, very impressive model in terms of bang for your buck, right? So, Deep Seek V4 Flash is not only incredibly fast, like some of these token throughputs are are crazy. Like, it's really, really fast. It also honestly is better than a lot of other models. Now, apparently what they've done is, this is just what I'm hearing, they've made it way more efficient for it to run this amount of compute, right?
Because they don't have a lot of compute over there, uh, apparently. Anyway, so yeah, for it to be number 17 in programming already, very, very interesting. This is a very, very popular model. I'm going to guess it's being used by um all kinds of AI agents and things like that. Okay, so this is very interesting. So top 11% of models. That's very very interesting. Code categories, top 16%. This is their flash model, guys. So yeah, it's being used in Hermes. I thought it might be Hermes and Open Claw. One thing that's missing is I don't think this is multimodal.
So, um, that's definitely interesting. I wish it was multimodal, but like apart from that, honestly, it seems like such a strong model for a flash model. This reminds me of when um Google first released what is it called? Gemini 3 flash. What is the pricing point on this? So, it's it's actually cheaper than Gemini 3 flash. So, that is very interesting. It's three times cheaper on input tokens and 10 times cheaper on output tokens. Really interesting. What about HighQ 4.5? I am actually waiting for HighQ 4.7 from Anthropic. It's kind of what I really really want them to come out with right now, but it doesn't seem like they are.
Instead, I'm going to be using a lot of Deep Seek V4 Flash um in the coming months because I really really like flash models. I've talked about this many, many times. It's got a million context window. It's stupid cheap. It's stupid fast. It works on open code. It works on CC router as well. So you can use it inside claw code router as well if you want to. Let's just see how it looks for front end design. And then I'm going to continue probably Oh, that is that is not bad. Wow. That is seriously not bad for a flash model, guys.
What the hell? So this is not only front end, it's also back end, right? I actually also um also do the dashboard in the same style. Like I can just keep adding to this as well. That's the craziest thing. This is super easy to set up. The prompt here was three lines, right? It was you have access to the Superbase CLI. I'm logged into the Superbase CLI. Make me a front end and a backend orth system using Superbase for a SEO blog writing project. Right? That was literally the prompt. And I mean, I'm impressed. I have to say this front end is very that that hero is really powerful and strong.
You can tell that they've been getting the output from Claude Opus, which is apparently what happened. Um, which is very interesting. You know, I'm down though, honestly, because we can't let Anthropic just be a complete monopoly on this stuff because they will abuse that 100%. Just as maybe a point of reference, this is what Opus 4.7 came up with for Harbor. Obviously, this is a slightly different project. But like I really like what Deep Seek V4 Flash has come out with here. Like literally in 15 minutes and for very very low cost. So 23 cents, I've got the bare bones of an entire project here.
Right. One thing that does keep happening is it keeps stopping my server, but that's okay. Okay, so this is what it's done for the dashboard, right? Really, really nice, I have to say. Honestly, guys, this is seeming like my new favorite model that's not Opus, right? And I'm not saying this is like Opus level, obviously. Um, but for the amount of coding power, the it made so few mistakes in creating this. It got a little bit stuck on the, you know, sign in sign out thing. I mean, that's pretty normal. It's not that easy for it to code that stuff.
I didn't give any skills either. So, I'd like to redo this test with skills. So, let's just try doing the actual uh writing part of this. So, I'm going to actually get Deepseek V4 Flash to power Deepseek V4 Flash here. So, I'm just going to grab this and just copy this and say now make the blog posting blog writing element. The user gives a website plus a keyword. Use scrape plus web search using open router web search it get up if you how and then generate a blog things you find. Right. I'm just going to give it it's a fairly broad prompt.
Uh wait I will set an open uh open key. So yeah look how fast this thing is guys. It's actually kind of insane how fast this is. This is why I love um these models, flash models. I really, really like Flash models for this reason. I want you to only use Deep Seek Flash. I mean, look at this thinking. Honestly, this is super super interesting. Wow. Use Cheerio passing. Sure. Okay, so this is what people actually want to see in my opinion. They don't want to see uh front end stuff only. Uh people will honestly would rather see backend stuff or I would anyway like it even has the foresight to change this button and things like that.
This is not very common with cheap models and especially not uh cheap Chinese models. So in my opinion, this does start to feel like a step forward for kind of open source. I'm hoping that people can learn from Deepseek because I mean as we know Deepseek is very very transparent like massively transparent and they wrote a paper on how they did this. Um yeah I'm not the right person to go through this. There's loads of videos on YouTube but um they've written detailed papers on how they managed to do this. It's something to do with the efficiency of the models, but it's very very exciting stuff.
And if people can learn from this, like Kimmy for example can come along and do what Deepseek have done, but you know, maybe slightly better in a way because you know, if you compare Kimmy to Deepseek, I would say Kimmy 2.6 is better than older Deep Seek models. So, I'm hoping that this is the start of something new for open source. Okay, so let's just do a quick run here. Okay, so it's running. Um, wow. Very, very interesting stuff that you can even get close to here. So, it said fail to connect here. Okay, so it looks like it might actually be running here.
I didn't change anything then. It was just um open code was still making changes. I mean, this is insane. You I don't think you guys understand like how impossible this was with these models until very very recently to get anywhere close to this. You'd be struggling for hours and hours and hours to be able to do this on a Chinese open-source flash model which costs nothing to run is absurd, honestly. Okay, so it's finished. It's written it. Um, yeah, I'm not seeing images, but this does look like it's got some information. Yeah, offers a bespoke kit on sneaker service.
This needs some changing around, you know, to get some images and stuff like that. But I mean, this is crazy to even to get to this point is crazy. Yeah, look, it's actually got pricing here. So, Kiton's black cashmre sneakers. I wonder if this does actually have links in. It's just not showing up. Let's just have a quick look. My theory might be right here. Let's see. No, there's no links. Okay, but still, it got real information from the website, right? So, that's the first step. All we need is a way to now get images and we've created an entire SAS project using only Deep Seek V4 Flash.
I think this is my new favorite model, guys, to play around with. I'm going to be making probably quite a few more videos about this. Um hopefully they won't go ahead and nerf this model because this does seem really really nice and really really strong. Looks like it did forget to um like seed the database with the blog posts that are generated, but that's just a very very small change. So yeah, we have a full signin sign out with literally the e most simple prompt you've ever seen. It was literally just you've got access to the superbase CLI.
Create me an orth project on Astro where I can sign in and sign out. That is done. It's done a fantastic job with the front end. I have to say really really nice job with the front end. Back end as well. Well, it managed to do that, too, which is crazy, guys. Okay, guys. So, I think I'm going to leave the video there. Just before I finish, quick shout out to harbor seo.ai, my actual blog writing SAS. Go and check it out, guys. I'll leave a link in the description and in the pin comment. I see that impressions and clicks are growing every single day.
The amount of pages that we've published are going up every single day. For €29 a month, you get 25 articles. You will not find a better SEO tool on the market for that price, guys. You get a free trial which gives you full access to the tool for a few days so you can decide whether or not you like it. And if you don't, you can cancel the trial. There is a link in the description and in the pin comment. With that being said, guys, thank you so much for watching. If you are watching all the way to the end of the video, you're an absolute legend.
And holy crap, things are about to get exciting with open source being able to produce something so amazing. I'll see you very very soon with some more content.
More from Income stream surfers
Get daily recaps from
Income stream surfers
AI-powered summaries delivered to your inbox. Save hours every week while staying fully informed.









