Foundation Model
A general-purpose model trained on broad data that can be adapted (via prompting or fine-tuning) to many downstream tasks.
Last updated: April 26, 2026
Definition
Foundation models are large neural networks trained on broad data at scale, designed to serve as the base layer for many specific applications without retraining. Claude Opus, GPT-5, Gemini 2.5, and Llama are foundation models. Their value comes from generality: the same base model handles summarization, coding, customer service, classification, and reasoning, adapted to each task via prompting (or in some cases, fine-tuning). The term originated in the Stanford CRFM 2021 paper that named the category. Production AI engineering today is mostly about adapting foundation models, not training them.
When To Use
Default to foundation models for any new application. Custom training is rarely worth the cost when an off-the-shelf foundation model can be adapted via prompting.
Related Terms
Building with Foundation Model?
I've shipped this pattern in real production systems. If you want a second pair of eyes on your architecture, that's what I do.