Foundation Model

Webisoft AI team Updated 2 min read

A foundation model is a large AI model trained on broad data at massive scale that serves as a general-purpose base, designed to be adapted, through prompting, fine-tuning, or additional tooling, to a wide range of downstream tasks rather than built for one job. The term (coined by Stanford researchers in 2021) captures the shift from training a bespoke model per task to building many applications on one shared foundation. GPT-series, Claude, Gemini, and Llama are all foundation models.

The relationship to nearby terms: a large language model is the most common kind of foundation model, but the category is wider, it also covers models trained on images, audio, code, protein structures, and combinations of these (multimodal models). “Frontier model” usually refers to the most capable foundation models at any moment.

The economics explain the ecosystem you see. Training a foundation model costs enormous compute, so only a handful of labs do it; everyone else builds on top, which is why thousands of AI products with different logos often run on the same few underlying models, differentiated by their prompts, data, and workflow around the model.

Why it matters at work

Understanding the foundation-model layer sharpens vendor evaluation. When a tool claims “our AI,” the useful questions are: which foundation model is underneath, what has the vendor added on top (retrieval, fine-tuning, workflow, guardrails), and what happens to your data at each layer. It also frames build-versus-buy honestly: your team will almost certainly never train a foundation model, but adapting one, with prompts, your documents, or fine-tuning, is well within reach of ordinary engineering budgets.

A work example

A product manager comparing three AI contract-review vendors discovers all three run on the same foundation model, so she scores them on retrieval quality, security posture, and workflow fit instead of taking “proprietary AI” claims at face value.

Get one practical AI-at-work idea in your inbox each week, subscribe to the AI Work+ newsletter.

FAQ

Is a foundation model the same as a large language model? Every modern LLM is a foundation model, but the category is broader. It also includes models trained on images, audio, and other data that serve as a general base for many downstream tasks.

Why are they called foundation models? Because one broadly trained model serves as the base for many applications. Teams adapt it with prompting, retrieval, or fine-tuning instead of training a new model for every task.