Embedding Model
The model whose whole job is to turn text into vectors that place similar meanings close together.
Think of it like
A translator that renders every sentence into the same universal coordinate system, so closeness means similarity.
Example
You run all your support docs through an embedding model once, store the vectors, and reuse them for every future search.
How it actually works
An embedding model is trained so that texts with similar meaning land near each other in vector space and unrelated ones land far apart. It’s usually separate from and smaller than the generation model, and its quality sets the ceiling for any retrieval built on it — a weak embedder means poor matches no matter how good everything downstream is. Choosing one involves tradeoffs of accuracy, vector size, speed, and language coverage.
For product teams
The embedding model quietly caps the quality of every search and RAG feature built on top of it — choose deliberately.
For engineers
A model mapping text to fixed-length vectors trained so semantic similarity corresponds to vector proximity.
Related
- Embeddings — The vectors it produces.
- Similarity Search — What those vectors get searched by.
- RAG — The system it feeds.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome