Getting Started
Understanding AI models
You've heard of ChatGPT. But what's a model — and what are GLM, Grok, or Sonnet? This is a plain-language guide to the thing that actually builds your site, the three models Ullbek supports, and how to choose between them. (Short version: you usually don't have to.)
What is a model?
ChatGPT is a product. The thing powering it — the brain that reads your words and decides what to say back — is a model. When you ask Ullbek to build a website, a model is what reads your request, writes the HTML and CSS, and makes the thousand small design decisions along the way.
Different companies build different models. OpenAI makes the GPT family; Anthropic makes the Claude family (Sonnet is one of them); xAI makes Grok; Zhipu makes GLM. They're all trained to do similar things, but they have genuinely different personalities and strengths — a bit like different people who happen to be very good at the same job.
That's the whole reason more than one exists. No single model is the best at everything. They differ in what they can perceive, how they reason, how fast they are, what they cost to run, and — yes — in taste. Two models given the exact same brief will build two different sites.
Why there's more than one
Ullbek supports three models out of the box. You can pick whichever you like — but you never have to. We pick a sensible default so that most people never think about this at all.
We expose the choice because some people like to try different models and settle on the one whose work they prefer. Over time you get to know a model the way you get to know a collaborator. But it's an option, not a chore. If you never touch the setting, you'll be fine.
The three models, side by side
Here are the three models Ullbek supports today, and what each one is good at. The capabilities that matter most for website building are text (reading and writing code and copy), vision (looking at images and screenshots), and PDF (reading documents you upload).
| Model | Text | Vision | Our take | |
|---|---|---|---|---|
| GLM 5.2 | Yes | — | — | The default. Excellent taste for website work at a gentle cost. |
| Grok 4.5 | Yes | Yes | Yes | Text and vision. Can look at screenshots and read PDFs. |
| Sonnet 4.6 | Yes | Yes | Yes | Text and vision. A different personality for the same capabilities. |
A little more on each:
Text only. GLM can't look at images or read PDFs — but it doesn't need to in order to build a beautiful site. It has strong design taste for website work, and it's the most economical of the three. That combination is why it's our default.
Text, vision, and PDF. Grok can look at an image or screenshot you share and reason about what it sees — useful when you want the model to match a reference, fix something visual, or read a document you've uploaded.
Text, vision, and PDF. Sonnet has the same capabilities as Grok but a different personality — it approaches problems in its own way. Some people come to prefer how it builds. The only way to know is to try it.
What "vision" means — and what it doesn't
Vision is the ability to literally look at an image. A vision model can see a screenshot you share, notice that a heading overlaps a photo, and fix it. That's genuinely useful.
But here's the thing people get wrong: a text-only model is not blind to your website. When something visual needs checking, a text-only model falls back to other mechanisms — it measures the layout mathematically (spacing, overlap, geometry) and reads the code directly to find what's wrong and put it right.
Vision model
Looks at the rendered page like a person would — spots a cropped headline or a clashing color by seeing it.
Text-only model
Reads the code and measures the layout — catches the same problem by inspecting the numbers and the structure behind it.
Why GLM is the default
We choose the default on a simple basis: price and performance. It's the model we feel is the best right now for building beautiful, performant websites at a reasonable cost — so most users don't burn through their credits faster than they need to.
GLM has excellent taste for website development specifically. For the vast majority of users, the default is simply enough. You don't need to think about models at all, and your sites will be good.
We revisit this as new models arrive. If a better price/performance option appears, the default moves with it — so staying on the default means you're always on what we believe is the best all-rounder for the price.
A model is like a person
Here's the part that surprises people: a model is less like a tool and more like a person. You get to know it. You notice the way it tends to lay out a page, the kind of spacing it reaches for, how it phrases things. Some people slowly develop a familiarity with one model and stick with it — not because it's objectively better, but because they like how it works.
There is no one-size-fits-all. Two capable models will build two different sites from the same brief, and neither is wrong. That's the whole reason we give you a model selection: it's a choice you can make, not a setting you must master.
How to choose
In practice, the decision is small. Here's the whole guide:
-
1
Most users: stay on the default
It's picked to be the best all-rounder for the price. You don't need to do anything.
-
2
Want to try a different personality? Switch and compare
Pick another model, build something, and see how it feels. You can always switch back. There's no penalty for experimenting.
-
3
Need it to look at a screenshot or PDF? Pick a vision model
If you're sharing an image you want matched, or a document you want read, choose Grok or Sonnet — they can see. The default can't.