What is Foundation Model? — AI Glossary

What it is: A foundation model is a large AI model trained on a huge, diverse dataset that can then be adapted to many specific tasks. ChatGPT, Claude, Gemini, and Llama are all foundation models.
Who it is for: Anyone who wants to understand the AI landscape in 2026 — the term is used constantly in industry conversations and reporting.
Best if: You want to understand why so many AI products feel related and what the term “foundation model” means in business reporting and AI strategy discussions.
Skip if: You just use AI tools casually and don’t care about the underlying architecture. Want one practical AI workflow every morning? Subscribe to our free daily newsletter.

What is a foundation model?

A foundation model is a large AI model that’s been trained on a huge, diverse dataset and can be adapted to many specific uses without retraining from scratch. The term was coined by Stanford’s Center for Research on Foundation Models in 2021 to describe the new class of models — like GPT-3, BERT, and CLIP — that didn’t fit cleanly into older AI categories.

What makes a model a “foundation”: scale (huge parameter counts and training data), generality (it can do many tasks rather than one), and adaptability (a single model can be specialized via prompting, fine-tuning, or RAG for downstream uses). Large language models are the most familiar foundation models, but the category also includes image models (DALL-E, Imagen), video models (Sora, Veo), and multimodal models (GPT-4o, Gemini, Claude).

Why does the term “foundation model” matter?

Before foundation models, AI products were built for narrow tasks: a spam filter, a translation system, a face recognizer. Each required its own dataset and its own training pipeline. The foundation-model paradigm flipped that: one extremely capable general-purpose model, adapted at the edges for specific uses.

That shift concentrates AI capability in a handful of organizations big enough to train foundation models — currently OpenAI, Anthropic, Google, Meta, xAI, Mistral, DeepSeek, and a few others. Everyone else builds on top. The term shows up in policy debates (should we regulate foundation models?), enterprise procurement (which foundation model do we standardize on?), and investment analysis.

How do foundation models compare to frontier models?

You’ll hear both terms. They overlap but aren’t identical:

• Foundation model — a general-purpose model trained at scale that can be adapted to many tasks. Includes everything from older models like BERT to current frontier ones.
• Frontier model — the most capable foundation models at the cutting edge in a given moment. GPT-5, Claude Sonnet 4.5, Gemini 3 Pro are frontier models in 2026.

In short: every frontier model is a foundation model, but not every foundation model is at the frontier. Frontier model is a moving target; foundation model is the category.

Related terms

• Frontier Model
• Large Language Model (LLM)
• Pre-training
• Fine-tuning

Learn more on Beginners in AI

• AI Glossary

Sources and further reading

• Stanford Center for Research on Foundation Models
• On the Opportunities and Risks of Foundation Models (Stanford, 2021)

Last reviewed: May 2026. AI terminology evolves quickly — verify specifics on the official source pages above.

You may also like

• AI Glossary
• Frontier Model
• Large Language Model (LLM)
• Pre-training
• Fine-tuning