---
language: en
title: "Foundation model"
description: "A foundation model is a base trained on broad data and adaptable to various tasks. Learn to identify dependencies and evaluate the application that uses it."
canonical: "https://tropk.ai/en/resources/glossary/foundation-model"
last_modified: "2026-10-12T16:13:39.602Z"
---

# Foundation model

Reviewed on 2026-10-06.

## What it means

A foundation model is a model trained on broad data at scale that can be adapted to different tasks. The term highlights its role as a base for downstream applications. This base can work with language, images, or other modalities. The concept does not describe a ready-to-use application or guarantee that every adaptation will perform adequately in its use context.

## Why it matters for marketing

When choosing tools, the team needs to recognize when different products depend on the same base. A foundation model failure may recur across several applications, although each adds its own instructions, data, and controls. Identifying this dependency helps the team ask questions about evaluation, version changes, and responsibility for errors in the content or analyses delivered to them.

## Example in practice

Consider an agency evaluating two services: one summarizes consumer research, and the other writes campaigns. The providers’ documentation lists the same foundation model, and hypothetical tests of both services omit caveats from a testimonial. The agency records the shared base and requests a specific evaluation of this failure in each application, without treating the two services as independent technology alternatives.

## How to apply or verify

Ask for the base model’s name and version, when disclosed, and a description of the adaptations made by the provider. Test real tasks using separate criteria for accuracy, instruction following, and preservation of caveats. Keep the inputs, outputs, and test conditions. If the application changes models, repeat the relevant cases before assuming its previous behavior will be preserved.

## Important distinctions

A foundation model can be generative, but the terms describe different properties: the adaptability of the base and the production of content. [Frontier model](/en/resources/glossary/frontier-model) refers to the relative position of capabilities at a given time. [Fine-tuning](/en/resources/glossary/fine-tuning) is one way to adapt parameters; providing documents during a query is not, by itself, this kind of training. The final application still includes its own implementation choices.

## Sources and references

- [On the Opportunities and Risks of Foundation Models](<https://arxiv.org/abs/2108.07258>) — Stanford CRFM / arXiv
- [Fine-tuning](<https://huggingface.co/docs/transformers/main/en/training>) — Hugging Face

## Continue a leitura

- [Fine-tuning](</en/resources/glossary/fine-tuning>)
- [Generative AI](</en/resources/glossary/generative-ai>)
- [Frontier model](</en/resources/glossary/frontier-model>)

## About the author

[Vinicius Pacheco](</en/team/vinicius-pacheco>)

Responsible for Tropk’s technical content, focusing on crawling, indexing, structured data, and agent access to content.

---

[All pages in Markdown](<https://tropk.ai/sitemap.md>)