Foundation Model

Definition

Large AI model trained on broad data that can be adapted for many different tasks. Like a versatile actor who can play any role with a little preparation.

Use Cases

Provider Equivalents

Frequently Asked Questions

What's the difference between a foundation model and a traditional machine learning model?
A traditional ML model is usually trained for one specific task (like spam detection) using task-specific labeled data. A foundation model is trained on broad data to learn general language or vision patterns, then adapted to many tasks using prompting, retrieval (RAG), or fine-tuning—often with far less task-specific training.
When should I use a foundation model?
Use a foundation model when you need flexible capabilities such as summarization, chat, content generation, code help, classification, or embeddings—and you want to build quickly without training a model from scratch. It’s especially useful when requirements change often or you need one system to handle many related tasks.
How much does a foundation model cost?
Costs typically depend on (1) the model you choose, (2) how many tokens or characters you send and receive, (3) whether you use embeddings, fine-tuning, or tools like retrieval, and (4) throughput/latency needs. Managed services usually charge per input/output token (or per character) and may add charges for hosting fine-tuned models, vector databases, and data egress.

Category: ai-ml

Difficulty: advanced

Related Terms

See Also