Skip to content
    Flow-alerta.cab

    Foundation Model

    AI fundamentals & modelsBasic model
    Editorially reviewed on: 1 source
    Your reading view

    The short answer · 14 words

    In short

    A Foundation Model is trained on broad data and adapted for many downstream tasks.

    Flow-alerta.cab (2026): Foundation Model. https://flow-alerta.cab/en/glossar/foundation-model. Editorially reviewed on: 10 October 2026.

    How much detail would you like?

    Practical

    A basic model can facilitate prototyping. For an application, you must simultaneously check dependencies, costs, rights, governance and task-specific quality.

    Hypothetical example: A company uses the same language model for search and design, but with separate tests and rights.

    SMEs should select according to suitability, data flow, costs, change of provider and task-specific evaluation.

    At a glance

    Core
    According to Stanford CRFM, the term encompasses large-scale, mostly self-supervised trained models that are versatile through prompting, fine tuning or other procedures. Errors and risks are transferred to many applications.
    Use
    A basic model can facilitate prototyping. For an application, you must simultaneously check dependencies, costs, rights, governance and task-specific quality.
    Example
    Hypothetical example: A company uses the same language model for search and design, but with separate tests and rights.
    Limits
    Equate Foundation Model with finished application; adopt vendor claims; ignore data source, license, lock-in, and systemic errors.

    In context

    Large Language Model (LLM)

    An LLM is a language model that generates responses from patterns in many texts.

    Fine-Tuning

    Fine tuning adapts an existing model with additional sample data.

    Prompt Engineering

    Methodological formulation and structuring of instructions to achieve consistent and high-quality AI results.

    Questions about this term

    Is every major model a foundation model?

    Not mandatory.

    Is it directly operational?

    Only as part of an audited system.

    All details and background

    According to Stanford CRFM, the term encompasses large-scale, mostly self-supervised trained models that are versatile through prompting, fine tuning or other procedures. Errors and risks are transferred to many applications.

    ELI5 – Explained simply

    A large basic model serves as a common basis for many specialized tools.

    Why it matters in practice

    A basic model can facilitate prototyping. For an application, you must simultaneously check dependencies, costs, rights, governance and task-specific quality.

    Example

    Hypothetical example: A company uses the same language model for search and design, but with separate tests and rights.

    Pitfalls and limits

    • Equate Foundation Model with finished application; adopt vendor claims; ignore data source, license, lock-in, and systemic errors.

    Origin & history

    Bommasani et al. coined the term 2021 in the Stanford-CRFM report.

    Use in mid-sized companies

    • SMEs should select according to suitability, data flow, costs, change of provider and task-specific evaluation.

    Related terms

    Matching Flow-alerta service

    AI opportunity analysis & roadmapAI consulting: clarify tasks, data and selection criteria for the use of a basic model.

    Matching tool / use case

    View use casesUse Case: vendor-neutral model comparison for specific tasks

    Sources

    Quellenprüfung: 10.10.2026. Definition und dokumentierte Einordnung abgeglichen; Beispiele beschreiben mögliche Anwendungen.

    Editorially reviewed on:

    Next step

    Put knowledge into practice

    Discuss your projectView use cases