Foundation Model

Short definition

A foundation model is a large AI model trained on broad data so it can support many tasks. It serves as a reusable base for chatbots, copilots, classifiers, search tools, image systems and domain-specific applications.

How it works

The model is first trained at scale, often using self-supervised learning. Teams then adapt it through prompting, retrieval, fine-tuning, tool use or product-specific layers. A large language model is one common type of foundation model.

Example

A company might use the same foundation model to power support summarization, internal search and document drafting. Each product can apply different prompts, permissions and retrieval sources while sharing the underlying model.

Why it matters

Foundation models shift AI from narrow one-task models toward reusable platforms. The benefit is speed and flexibility. The risk is dependency on a model whose behavior, training data and limits may not be fully transparent.

A foundation model is not a finished product

The application still adds instructions, an interface, data sources, permissions, moderation, monitoring and operational controls. Two products using the same underlying model can therefore have very different quality and risk profiles.

How to evaluate a model choice

Look beyond public benchmarks. Review the license, supported regions, data policy, API stability, inference cost and the effort required to change providers. Internal evaluations should reflect the languages and tasks of actual users. The highest general benchmark score may not identify the best model for classification, legal documents or on-device use.