What Is An AI Model?

Psst! Before reading this post, I highly recommend checking out my previous article on “What Is AI?” for a basic overview.

In today’s episode of “things I pretended to understand for a while even though I didn’t”, I present AI models.

AI models are things that tend to get thrown around a lot by people who want to remind you that you know NOTHING about AI:

“oh are you using GPT-4o or GPT-mini?”

“I prefer Claude Sonnet myself…”

Huh??

So in this blog, I want to go through exactly what an AI model is, how they are created, why they are important and where they are used.

Basically, everything you ever wanted to know about AI models.

Let’s get started:

If you have read my post on What is AI?, you will know that AI is a big umbrella term that can mean a lot of stuff: a product, a technology, a method.

But where does the AI model fit in?

An AI model is a computer program that has been trained on large amounts of data to recognise patterns and make predictions or generate outputs based on those patterns.

In other words, at the heart of every AI tool, product or technology, there is an AI model (a computer program) which is powering the process.

For example:

When we interact with tools like ChatGPT, Gemini or Claude, really we are interacting with an AI Model.

ChatGPT is just a product that was built on top of the AI model, to make it easy for you to talk to it.

Side note: I personally like to think of an AI Model as a sort of "brain", but if you like to keep your humans/computer life nice and separate, you can consider it the “engine” of any AI tool or product you might have come across.

An important thing to know is that there are many different AI models.

AI models are constantly being developed and improved upon, at a rapid rate!

Huge AI companies are investing literally billions to create the smartest, fastest and all-round best AI model there is.

Why?

Because the smarter the model, the better you can solve problems using it, the better the tools you can build around it, the more money you can make from it.

(sorry to disappoint, but yeah, turns out AI is mostly just about money)

This is why there are lots of different AI models on the market right now.

As of writing this in May 2026, here are a few major ones:

Don’t worry, you don’t need to remember these. Just remember that your AI tool of choice is being powered by one of these. Open up your favourite AI tool and see if you can see which model is being used to power it.

Before we carry on, a quick reminder on the difference between an AI Product and an AI Model.

(because this was something that confused me at first)

So if AI is the what then the AI Model is the how.

With me so far?

Then let’s go one step deeper and look at how AI models are actually created:

AI models are created by very smart people, usually working for big tech corporations (not me sadly).

To look at how AI models are created, it’s worth remembering how things used to be before AI came along…

Computer programs (lines of code that computers can read), have been around for a long time.

Previously, smart people would write “commands” or “rules” that a computer could execute line after line (if this, do this, then do this…)

All of these commands running together to complete a task = a computer program.

But AI computer programs (AI models) can’t work like this…

A classic computer program follows a strict set of rules, but for AI to work, we need it to be able to adapt, adjust, figure things out for itself, without being told how to (like a human would)

We needed to create a totally different type of computer program.

So how did we solve this problem?

Instead of a list of rules (if this, do this), we give a computer program a huge bunch of training data - an enormous amount of real life examples - and let it figure out the patterns for itself.

You can imagine giving a child an enormous stack of photos of cats and dogs.

You give the child a few examples, then ask it to figure out the rest of the photos on its own.

The child looks at the photos and sees “hmm, cat’s have whiskers”, “dogs are usually bigger”, “cats have pointy ears”…

You never strictly tell the child any of this. It just figures it out through looking at examples.

Now when you give the child a new photo and ask it “is this a cat or a dog?” It can think about all the previous examples it has seen and can answer “cat” or “dog” with a pretty high chance of being correct.

This training process is called machine learning and is how AI models are created.

Roughly speaking, we can split it into 3 steps*

The model that comes out of all this training is a piece of software called an AI model.

*how AI models are created can become a very technical topic very quickly. This is a rough summary but please see my post on “machine learning” for more details.

Very good question!

There are so many AI models out there, how do I know which model to use?

Well, that would depend on a few different things:

As I said before, different AI companies have created different AI models.

You don’t need to tell it which model to work with, it chooses this automatically. You can see this usually by looking at the tool:

People tend to have a (sometimes irrational) preference for the tool they like to use.

Me personally, I am team Claude because I agree with their ethics and principles more than other big AI corporations.

But perhaps you work in an office which strictly wants you to use Gemini or ChatGPT…

Maybe you are a total AI nerd and heard that Gemini just released an AMAZING new AI model and you definitely want to use that.

The company you want to interact with (and potentially pay money to) will have an impact of the choice of AI model you use.

I already mentioned that there are many different AI models out there.

One reason is that companies are competing with each other to create the BEST AI model there is (bigger, faster, more data…)

But another reason is that we can program different AI models which are good and doing different things.

For example:

You should pick the model which has been optimized for the task you are looking to do.

If you are specifically generating images, you should google “which AI model is best at generating images” and play around with each to find the one that works best for you.

Why is it that everything comes back to money?

Of course this is a huge factor for deciding which AI product (and thereby model) to use.

Gemini, Claude and ChatGPT all have different pricing options depending on which plan you want to use.

You might therefore decide “arggggghhhhh Gemini is WAY too expensive, I’m switching to Claude” and therefore to a different AI model.

This is ultimately the reason that AI companies are racing to produce the bigger and better AI model. The better the model…the more they can charge for it.

Capitalism eh.

The question that should be at the front of everyone’s mind when using tools that are built using AI models.

How accurate are they really??

Well, if you remember only one thing, remember this:

It is IMPOSSIBLE for an AI model to be 100% accurate about ANYTHING.

I know, that seems weird right? If you ask Gemini “what is 2+2?”, you would imagine that there is a 100% of it always returning “4”.

But this is not true.

The likelihood is maybe something like 99.999999999% but NEVER 100%.

This is because an AI model is not a “solution provider”.

It is a “prediction Wizard”.

AI models work by predicting what the most statistically likely answer will be, given all of the examples it has ever been trained on. It does this very, very fast, over and over again, until it has produced a full response.

This is why models can sound brilliantly confident and still be SO wrong.

They are giving you the most probable answer based on their training data, and sometimes, that answer is just... wrong.

This is called hallucination.

When a model produces an answer that sounds plausible and is stated with total confidence (like, a lot! of confidence), but is factually wrong.

This comes back to the point I made earlier about bad training data = bad model. If the data used to train the model was poor quality, biased or incomplete, the model will produce low accuracy answers.

So remember before you let AI write your wedding vows, or skim read an important contract for you - AI is never 100% sure of anything!

CHECK YOUR ANSWERS!

The most important bits from this post: