How to Build a $2,000 Website for $2.42 With AI

A YouTuber built a full professional website last week. Custom graphics. Responsive design. SEO-ready copy. Total cost: $2.42.

Not $2,420. Two dollars and forty-two cents.

The client he built it for would have paid $1,500 to $3,000 for the same thing from a freelancer. The AI did it in 45 minutes. And right now, there's a narrow window where this specific trick works at this price point.

We're not talking about replacing web designers. A pricing arbitrage exists right now because a new AI model just dropped at absurdly low rates. The window to use it at this price is closing fast.

Why This Works Right Now

A Chinese company called Z.ai released a model called GLM-5 in early February 2026. It's a 745-billion parameter model that scores just behind Claude Opus 4.6 on benchmarks. The twist: it costs roughly $1 per million tokens. That's 90% cheaper than the top American models.

For context, the previous cheapest option that could build a decent website was Google's Gemini Flash. It's fast but the designs look generic. GLM-5 produces what one creator called "Opus-level" designs, meaning top-tier visual quality, custom SVG graphics, and thoughtful color schemes.

The catch? This pricing probably won't last. When DeepSeek launched their cheap model last year, the price went up within weeks as servers got hammered. Multiple creators are warning that GLM-5 will either get rate-limited or the quality will be "nerfed" (intentionally reduced) once demand overwhelms their infrastructure.

That's what creates the opportunity. Right now, you can build websites that look like they cost thousands, for the price of a cheap coffee.

The Tool Stack (No Coding Required)

You need two things:

1. Kilo Code is an AI coding agent. You describe what you want in plain English, and it writes the code, creates the files, and builds the entire site. Think of it like hiring a developer who works for pennies and never complains about scope creep.

2. GLM-5 via OpenRouter is the AI model that powers the design decisions. OpenRouter is basically a menu of AI models. You pick GLM-5, plug in your API key, and Kilo Code uses it to generate everything.

Optional but useful: Claude (any plan) for fixing bugs after the initial build. More on that in a minute.

Total setup time: about 15 minutes if you've never used these tools before.

Step-by-Step: Building Your First Site

The process breaks into five steps anyone can follow.

Step 1: Get your API key. Go to openrouter.ai, create a free account, and add $5 in credits. This will cover multiple website builds. Find GLM-5 in the model list and copy your API key.

Step 2: Set up Kilo Code. Install the VS Code extension (search "Kilo Code" in extensions). In settings, paste your OpenRouter API key and select GLM-5 as your model.

Step 3: Write your prompt. This is the part that matters most. Describe the website like you're briefing a designer:

"Build a modern, professional website for a plumbing company called Smith's Plumbing in Austin, TX. Include a hero section with a call-to-action button, services page, about section, contact form, and testimonials. Use a dark blue and white color scheme. Make it mobile-responsive. Include custom SVG icons for each service."

Step 4: Wait 45 minutes. GLM-5 is slower than Google's models (which finish in 7 minutes). But the quality difference is massive. You'll see Kilo Code creating files, generating graphics, and assembling the site in real time.

Step 5: Fix the small stuff. The site will look great but might have a few broken links or minor layout issues. This is normal. Use Claude or ChatGPT to fix specific bugs: "This navigation link goes to a 404. Fix it." Takes 5 minutes.

That's it. You now have a professional website that would cost $1,500+ from a freelancer.

The Economics: Why This Is Worth Your Time

Let's do the math on a single website sale.

Your costs:

• GLM-5 API tokens: ~$2.42
• Your time: ~1 hour (including setup and fixes)
• Hosting for client demo: $0 (use Vercel or Netlify free tier)

What you charge:

• Low end: $500 for a simple small business site
• Mid range: $1,000 to $1,500 for a multi-page site with custom graphics
• High end: $2,000 to $3,000 if you add SEO setup and content

Even at the low end ($500), that's a 99.5% profit margin on a single hour of work. Build three sites in a week and you're looking at $1,500 to $4,500 in extra income.

Compare that to traditional freelancing where you'd spend 20+ hours on a single site, deal with client revisions, and still charge the same amount.

Selling and Scaling Your Sites

Building the website is the easy part. Selling it is where most people quit. A few approaches actually work for finding clients who'll pay for this.

Target local service businesses. Plumbers, electricians, HVAC companies, landscapers, dentists. These businesses often have terrible websites (or none at all) and make enough money to easily afford $500 to $2,000 for a professional site.

There's a stat from marketing research that keeps coming up: the first vendor to respond to an inquiry wins 35% of the business. Local service companies know this. They just don't have the tech skills to fix it.

The demo-first approach works best. Don't cold-email asking if they want a website. Instead, build one for them first (it costs you $2.42, remember), then send them a message:

"Hey, I noticed your current website hasn't been updated in a while. I built a mockup of what a modern version could look like. No obligation, just wanted to show you what's possible. Here's the link."

This works because seeing is believing. A generic pitch gets ignored. A live demo of their own business, with their name and services, gets responses.

Where to find prospects:

• Google Maps: search "plumber [your city]" and look for businesses with no website or outdated ones
• Yelp: filter by businesses with good reviews but bad web presence
• Facebook local business groups: offer a free website audit
• Nextdoor: post about local web services

The real money isn't in selling a single $1,000 website. It's in turning each client into recurring revenue.

The maintenance package. After delivering the site, offer a monthly plan for $50 to $200/month that includes hosting, updates, and minor changes. Most business owners gladly pay this because they never want to touch their website again.

The SEO upsell. Once the site is live, offer search engine optimization as an add-on. Tools like Harbor SEO (mentioned by the same creator who demonstrated the $2.42 build) can automate keyword research and blog writing. Charge $300 to $500/month for "ongoing SEO management" that the AI mostly handles.

The template factory. After building 5 to 10 sites for different business types, you'll have a library of prompts that produce great results every time. A plumber prompt. A dentist prompt. A landscaper prompt. Each new client in that category takes 20 minutes instead of 45 because you already know what works.

At 10 clients paying $100/month in maintenance, that's $1,000/month in recurring revenue on top of the one-time build fees. And the AI does most of the actual work.

The Honest Tradeoffs

This approach isn't perfect. Be aware of these limitations before you start.

GLM-5 is slow. 45 minutes per build versus 7 minutes with Google's models. If speed matters more than design quality, use Gemini Flash instead (the designs just won't be as impressive).

The pricing window may close. Z.ai could raise prices, add rate limits, or reduce quality at any point. This is why acting now matters. Even if the model gets more expensive later, the skill of using AI coding agents to build sites transfers to whatever model comes next.

You still need to fix bugs. AI-generated code isn't perfect. Expect broken links, occasional layout glitches, and forms that don't actually submit. Budget an extra 15 to 30 minutes per site for cleanup using Claude or manual fixes.

Client expectations need managing. These sites look great but they're not custom-coded web applications. They're fast, attractive, functional sites for businesses that need a web presence. If a client wants complex features like booking systems or e-commerce, you'll need additional tools or a more experienced developer.

Sales skill matters more than tech skill. Greg Isenberg made this point in a recent video about using Claude Code for marketing: spend 50% of your time selling, not building. The AI handles the building. Your job is finding people who'll pay for the result.

What to Do Today

This website trick is just one example of a larger shift. AI models are getting better while simultaneously getting cheaper. The gap between what a solo person can produce and what a full agency delivers is shrinking fast. The playbook is simple: find something AI does cheaply, find people who'd pay real money for the result, and be the person in the middle.

For websites, that starts with five actions:

1. Set up your tools (15 min): Create an OpenRouter account, add $5 credits, install Kilo Code in VS Code, connect GLM-5.
2. Build a demo site (45 min): Pick a local business type you're familiar with. Write the prompt. Let the AI work. Fix any bugs.
3. Find 5 prospects (20 min): Search Google Maps for businesses in your area with bad or missing websites. Save their contact info.
4. Send demo-first pitches (30 min): Build a quick mockup for each prospect (remember, each one costs about $2.42). Send personalized outreach with a link to their demo site.
5. Close your first sale: Even at $500, your first client pays for months of API credits and proves the model works.

The window on GLM-5's current pricing won't stay open forever. But the skill you'll build, using AI to create things people will pay for, that compounds regardless of which model is cheapest next month.

Start with one website. See what happens.