Core
Foundation Model
A large-scale model trained on broad data that can be adapted to a wide range of downstream tasks through fine-tuning, prompting, or other techniques.
Explained at five levels
Level 1
A giant AI model that can do lots of different things — like a super-talented friend who's good at everything.
Level 2
A big, general-purpose AI model (like GPT-4 or Claude) that serves as the base for many different applications — from chatbots to code assistants to image generators.
Level 3
A large-scale model trained on broad data that can be adapted to a wide range of downstream tasks through fine-tuning, prompting, or other techniques.
Level 4
A model trained at scale on broad data using self-supervision, serving as a general-purpose base that can be adapted to diverse tasks via fine-tuning, prompting, or retrieval augmentation.
Level 5
A self-supervised model trained on internet-scale data that exhibits broad capability transfer — embodying the foundation model paradigm where a single pre-training run amortizes the cost of learning representations applicable across a distribution of downstream tasks.
Definitions are educational summaries. Terminology can vary by source and context.
Sources
- NIST Trustworthy and Responsible AI Resource Center: Glossary — National Institute of Standards and Technology. Accessed 2026-07-20.