Mehmet EmrahKonyaTR
← Back to glossary

Foundation model

Temel model

D2

In generative AI, a foundation model is trained on broad data using self-supervised learning and can be adapted for multiple application tasks.

Review status: 2026-11-26

Technical explanation

After pre-training, a developer can adapt a foundation model for a specific use case through further fine-tuning.

Conceptual boundaries

A foundation model is not necessarily an LLM, because foundation-model pre-training and adaptation also occur for image, audio, and multimodal models.

Provider-neutral example

A team can start with a broadly trained model and adapt it to classify internal document types for a defined workflow.

Limitations

This definition sets no parameter-count threshold and does not say that every model in the listed modality categories is a foundation model.

Related concepts

Atomic claims and evidence

  1. 1.1NIST AI 100-2e2025, Adversarial Machine Learning: A Taxonomy and Terminology of Attacks and Mitigations
    Source
    NIST AI 100-2e2025, Adversarial Machine Learning: A Taxonomy and Terminology of Attacks and Mitigations
    Source role
    Authoritative source
    Exact locator
    Appendix A p. 108: foundation model
    Supported claim
    In generative AI, a foundation model is trained on broad data using self-supervised learning and can be adapted, such as through fine-tuning, for a variety of application tasks.
    Last verification
    Review due
    Scope limitation
    This is NIST’s glossary definition and does not set a parameter-count threshold.
  2. 2.1NIST AI 100-2e2025, Adversarial Machine Learning: A Taxonomy and Terminology of Attacks and Mitigations
    Source
    NIST AI 100-2e2025, Adversarial Machine Learning: A Taxonomy and Terminology of Attacks and Mitigations
    Source role
    Authoritative source
    Exact locator
    Section 3.1.1 p. 37: training stage for generative image, text, audio, and multimodal models
    Supported claim
    Foundation-model training patterns occur for generative image, text, audio, and multimodal models.
    Last verification
    Review due
    Scope limitation
    The claim identifies listed model categories and does not say that every model in each category is a foundation model.