What Is an AI Model? (And Why Are There So Many?)

You hear it constantly now: “the new model is out,” “which model are you using?” GPT, Claude, Gemini, Llama, and a new one seemingly every week. But what is a model, really? And why are there so many?

Training: culinary school

A model starts out knowing nothing. During training, it studies an enormous amount of examples, like a chef-in-training cooking thousands of dishes. For a text model, that means reading a huge amount of writing and learning the patterns of language. (That’s the large language model kind.)

When training ends, what’s left is the model: all that practice boiled down into a giant file of numbers. It doesn’t keep the textbooks. It keeps the skills.

What makes one model different from another

• What it trained on. A chef trained on French cooking won’t make great sushi. Some models specialize in text, others in images, voices, or code.
• Size. Bigger models are like head chefs with decades of experience: better at hard problems, but slower and more expensive to run. Smaller models are quick line cooks, perfect for simple orders.
• Knowledge cutoff. A model stops learning when training ends. Ask about something that happened after that, and it simply doesn’t know, unless it’s allowed to look things up.
• Memory for your conversation. Some models can hold a whole book in mind at once. Others lose track after a few pages. This is often called the “context window.”
• Personality and rules. Companies tune their models to behave differently: more cautious, more creative, more concise.
• Open or closed. Some companies publish their models for anyone to download and run, like an open source recipe. Others only let you use them through their app.

Model vs. app

Here’s a handy distinction: the model is the chef, the app is the restaurant. ChatGPT and Claude are apps, the dining room you walk into. Behind the counter, they can swap in newer, better chefs over time without changing the sign out front.