Foundation models are trained on large general corpora and then adapted through prompting, retrieval, or fine-tuning to particular applications. The same base model can support drafting, summarization, classification, and question answering.
Most commercial legal AI products build on foundation models developed by a small number of providers.
Alternative Names:
Base Model, Pretrained Model
Why it Matters?
Because most legal AI vendors build on the same handful of foundation models, the model itself is rarely the differentiator. What distinguishes products is the surrounding engineering: document processing, retrieval quality, citation architecture, permissions, and workflow. It also means the underlying model provider is a subprocessor whose terms determine whether the vendor's confidentiality commitments actually hold.
Frequently Confused with
Related terms
Frequently asked questions
Do legal AI vendors build their own models?
Why does the underlying model provider matter?
