Foundation Model

Simple Definition

A foundation model is a large AI model trained on a huge and diverse dataset that can be used as a starting point for many different tasks. Instead of building a separate model for every task, developers build one large model and then adapt it.

GPT-4, Claude 3, Gemini, and Llama are all examples of foundation models.

Why “Foundation”?

The name comes from the idea that these models serve as a foundation — you build on top of them rather than starting from scratch for every application.

Before foundation models, most AI systems were trained for a single, narrow task. Now, one large model can handle writing, coding, summarization, translation, and analysis — and be adapted with minimal additional training for specialized uses.

How Foundation Models Are Built

  1. Pre-training — the model is trained on enormous amounts of text, images, or other data
  2. Fine-tuning — the pre-trained model is adapted for specific tasks or behaviors
  3. Deployment — the model is made available via apps or an API

Foundation Models vs. Task-Specific Models

Foundation ModelTask-Specific Model
Trained on diverse dataTrained on one type of data
Adaptable to many tasksGood at one thing
Expensive to train, cheap to adaptCan be cheaper for narrow uses
GPT-4, Claude, GeminiAn older spam filter or sentiment classifier
  • LLM — language-focused foundation models
  • Fine-Tuning — adapting a foundation model for a specific task
  • Generative AI — most generative AI tools are built on foundation models
  • Deep Learning — the technique used to train foundation models

See AI terms in action

Browse practical AI workflows that use the concepts in this glossary.

Last updated: