What Is a Foundation Model?
The term behind GPT, Claude, Gemini, and Llama — a single model trained once, then adapted for a huge range of downstream uses.
One model, many uses
A foundation model is a large model — usually an LLM — trained on broad, general-purpose data, meant to serve as the base for many different downstream applications rather than being built for one narrow task. The same underlying model can power a chatbot, a coding assistant, and a summarization tool, without retraining from scratch for each.
Why "foundation" is the right word
The name reflects the role: everything else gets built on top of it, through prompting, fine-tuning, or RAG. The foundation model itself rarely changes; what sits on top of it is where most product-specific customization happens.
Foundation models vs. narrow AI
Before foundation models became standard, most AI systems were purpose-built and trained end-to-end for one specific task, from scratch. Foundation models flipped that — train once at massive scale, then adapt cheaply and quickly for whatever's needed, which is a large part of why AI products multiplied so fast once they became available.
Frequently Asked Questions
What is a foundation model?
A large, general-purpose model — usually an LLM — trained once on broad data and then adapted, via prompting, fine-tuning, or RAG, to power a wide range of downstream applications.
Are GPT, Claude, and Gemini foundation models?
Yes — each is a foundation model developed by its respective provider (OpenAI, Anthropic, Google), which then gets adapted into specific products and features.