What Is a Foundation Model?

Models & Architecture

A foundation model is a large AI model trained on broad data that serves as a general-purpose base for many tasks. Rather than being built for one job, it learns wide patterns and is then adapted, through prompting or fine-tuning, to specific uses, which is why so many applications are built on top of a few foundation models.

How a foundation model works

A foundation model is trained once, at large scale, on vast and varied data, learning general patterns of language, images, or other content. This pretraining is expensive and done by a small number of organizations. Once it exists, the same model can be pointed at many different tasks without starting over.

Teams adapt a foundation model in a few ways: prompting it with instructions, giving it examples, connecting it to tools and data, or fine-tuning it on a narrower dataset. The large language models behind modern chat and writing tools are foundation models, as are the big image and speech models. The idea is a shared base that many specific systems are built on.

Why foundation models matter for AI

Foundation models changed how AI gets built. Instead of training a fresh model for every task, teams start from a capable general model and adapt it, which cuts the cost and time to build useful systems sharply. It also concentrates a great deal of capability, and a great deal of influence, in the few models most applications depend on. At Custom AI Studio, we build on top of existing foundation models rather than training from scratch, putting the customization into the data, workflow, and integration that fit a client's business.

Frequently asked questions.

The stuff we hear most on the first call. Don't see yours? Book a 30-minute conversation.

What is the difference between a foundation model and a large language model?

A large language model is a foundation model for text. "Foundation model" is the broader term that also covers image, audio, and multimodal base models. Every large language model is a foundation model, but not every foundation model works on text.

What are examples of foundation models?

The large language models behind popular chat and writing tools, along with major image, speech, and multimodal models. Each is a general base that many downstream applications adapt to their own tasks.

Do you need to train a foundation model yourself?

Almost never. Training one requires enormous data and compute and is done by a handful of organizations. Most teams use an existing foundation model and adapt it through prompting, retrieval, or fine-tuning.

Want to put AI
to work?

We work with leadership teams to find the right opportunities, define the strategy, and build the systems that move the business forward.

Explore the Workshop Discuss a Custom AI Project