Embedding model
Also known as: embedder
An embedding model is a model specialized to produce embeddings — numeric meaning-vectors — from text.
It’s usually smaller and cheaper than a full LLM because it does one job: turn text into a vector.
- Input: a sentence or chunk.
- Output: a fixed-length vector (e.g., 768–3072 numbers).
- Used by: RAG, semantic search, clustering.
The RAG workhorse: your document search quality is largely determined by the embedding model. It's the quiet component that makes or breaks retrieval.