Pretrained Model
Last Updated: July 29, 2026 | By Mihail Sebastian | AI Dictionary
A model already trained on a large general dataset, used as the starting point for new tasks instead of building and training a new model from scratch.
What is a Pretrained Model?
A pretrained model is a machine learning model that has already been trained on a large general dataset and is used as the starting point for new tasks instead of building one from scratch. Modern AI practice rests on this division of labor: a few organizations spend enormous compute training general models once, and everyone else adapts them.
GPT, BERT, and ImageNet-trained vision networks are the canonical examples.
How Pretrained Models Work
Pretraining exposes a model to broad data (billions of words, millions of images) so it learns general representations: grammar and world knowledge in language models, edges and textures and shapes in vision models. That work is done once and shared, through hubs such as Hugging Face.
Teams then adapt the model in three ways. They prompt it as-is with an instruction alone (zero-shot), show it a few worked examples in the prompt (few-shot), or continue its training on their own data (fine-tuning), the standard route of transfer learning.
The choice turns on data and stakes. Prompting costs nothing to set up; fine-tuning buys deeper specialization at the price of a training run and a dataset.
A pretrained model imports more than skill. Bias and gaps in the pretraining data carry into every system built on top, and downstream teams rarely know what that data contained.
Example of a Pretrained Model
A wildlife charity wants to identify species in camera-trap photos. Training a vision model from scratch would demand millions of labeled images the charity does not have.
Instead, it downloads a network pretrained on ImageNet. That model’s early layers already detect edges, textures, and animal shapes, so the charity fine-tunes it with a few thousand labeled camera-trap photos. The finished classifier reaches useful accuracy with a fraction of the data and compute.
Related AI terms: Fine-tuning · Transfer Learning · Zero-shot Learning · Deep Learning
Did you like the Pretrained 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