Model

What a Model Is

In machine learning, a model is the trained artifact that a learning algorithm produces. You feed data into a training process, that process adjusts millions (or billions) of numerical parameters, and what comes out is a model — a system that can take new inputs and produce useful outputs.

A model is defined by two things: its architecture (the design of the neural network) and its weights (the specific numbers the training process learned). The weights are what make each model unique.

Model vs Product

One of the most common confusions for beginners:

• GPT-4 is the model — a trained artifact created by OpenAI
• ChatGPT is the product — a chat interface OpenAI built on top of GPT-4

Similarly, Claude is Anthropic's model; Claude.ai is the product. The model is the AI itself. The product is the user-facing application built around it.

When you call an AI API from your code, you're calling the model directly — without the product wrapper.

Model Sizes and Tradeoffs

Models come in many sizes, typically measured in parameters (the numerical weights). Bigger models are generally more capable but also:

• Slower to generate responses
• More expensive to run (higher API costs)
• Require more compute to host

This is why AI providers offer model families — smaller, faster, cheaper models for simple tasks and large, powerful models for complex reasoning. Choosing the right model size for a task is one of the key engineering decisions in building AI applications.