Model
Last Updated: July 29, 2026 | By Mihail Sebastian | AI Dictionary
The trained artifact of machine learning: an architecture filled in with weights learned from data, packaged to turn new inputs into predictions or decisions.
What is a Model?
A model is the trained artifact of machine learning: a model architecture filled in with weights learned from data, packaged as a program that turns new inputs into predictions or decisions.
The model, not the training procedure, is what an organization deploys, sells, and answers for. (The everyday senses of the word, a business model or “modeling” a scenario, are not what this dictionary means.)
How a Model Works
A model begins as an architecture with random weights that predicts nothing useful. Model training feeds it data and adjusts the weights until its predictions match reality; everything the finished model “knows” lives in those numbers.
Model evaluation then tests it on data it never saw during training, because performance on memorized examples proves nothing. If the results hold up, the team will deploy it, and from that point it runs inference: answering live inputs, one prediction at a time.
The lifecycle does not end at deployment. The world drifts away from the training data, so deployed models need monitoring and, eventually, retraining.
Types of Models
- Supervised learning models learn from labeled examples to predict outputs for new inputs, as decision trees and neural networks do.
- Unsupervised learning models find structure in unlabeled data, as in clustering and dimensionality reduction.
- Reinforcement learning models learn by acting in an environment and adjusting to rewards and penalties.
- Generative models produce new data that resembles their training data: text, images, audio, code.
Model vs Algorithm
The algorithm is the recipe; the model is what the recipe produces. Gradient descent is an algorithm printed in every textbook, while the trained network it leaves behind is a model, and running the same algorithm on different data yields a different model.
The distinction carries legal weight. When a regulator examines “the algorithm” behind a lending decision, what actually needs inspecting is the model and the data it learned from.
Example of a Model
A team builds a classifier to tell cat photos from dog photos. They pick a convolutional neural network architecture, then train it on thousands of labeled photos; each mistake nudges the weights until the network reliably separates the two.
Evaluation on a held-out set of photos shows 96% accuracy, so the team deploys the model inside a photo app. A user uploads a new picture, the model runs inference, and out comes “cat, 0.98.” The file doing that work, architecture plus learned weights, is the model.
Related AI terms: Algorithm · Model Architecture · Weights · Model Training · Model Evaluation
Did you like the Model gist?
Learn about 250+ need-to-know artificial intelligence terms in the AI Dictionary.
Mihail Sebastian — Writes about AI governance, regulation, and the technology behind them. Placeholder bio — replace with a real credential line. About