Foundation Model

A foundation model is a big, broadly capable AI that acts as the groundwork for lots of other applications. Instead of training a fresh model from scratch for every new job, companies train one large model on an enormous range of data, then adapt it to particular purposes. The name captures the idea well: it is the foundation you build on.

The best-known examples are the large language models behind tools like ChatGPT and Claude, though foundation models exist for images, audio, and other kinds of data too. What they share is generality. A single foundation model can be pointed at translation, summarising, coding, drafting, and much more, without being rebuilt for each one.

The practical benefit of this approach is reach. Because the hard, expensive training is done once, a smaller team can take a foundation model and shape it for their own needs through fine-tuning or careful prompting, rather than starting from nothing. That is a large part of why AI tools have appeared so quickly across so many areas.

For you, the term mostly helps when you read about how AI products are made. When a company says its app is “built on” a foundation model, it means the heavy lifting came from a general base model, which they then adapted for their particular use.