Foundation Model — AI Glossary

Reference details

Related terms

• Large Language Model
• Pretraining
• Fine-Tuning

How it works

1. One model is pretrained on broad data at scale, rather than many models trained per task.
2. Capability is general enough that downstream tasks are reached by adaptation — prompting, retrieval, fine-tuning — instead of retraining.
3. The term is modality-agnostic by design: vision, audio and multimodal models qualify on the same criteria as text models.
4. Because many applications inherit from one model, its limitations and biases propagate to all of them, which is the concentration risk the term was coined to name.
5. Adaptation cost is small relative to pretraining cost, which is what shifted the economics from per-task training to per-task prompting.

Example: Why the economics favour adaptation

Training a task-specific model needs labelled data, a training run and a maintenance commitment per task. Ten tasks is ten of each. Adapting one foundation model is ten prompts, or ten small fine-tunes, against one pretraining cost nobody in the chain paid again. The arithmetic is why the field reorganised around adaptation — and the same arithmetic is the concentration risk, because a flaw in the shared model is a flaw in all ten applications at once.

Commonly confused with

Foundation model describes breadth and adaptability — trained on broad data, useful across many downstream tasks, and not restricted to text. Base model describes a training stage: pretrained but not yet instruction-tuned. They are different axes, so an instruction-tuned assistant is still a foundation model and is no longer a base model.

When to use it

Reach for it when:

• Describing broadly capable models adapted to many tasks, across any modality.
• Explaining why applications now start from a pretrained model rather than a training set.
• Discussing concentration risk, where one model's limitations propagate downstream.

Reach for something else when:

• As a synonym for large language model, which is one modality within the category.
• As a synonym for base model, which describes training stage rather than breadth.
• For narrow task-specific models, however large.

Continue your research

Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to Foundation Model — AI Glossary.

• Browse models
• Review providers
• Inspect benchmarks
• Explore workflows
• Learn AI concepts
• Read decision guides
• Review trust and methodology
• Start a decision path