Embedding Model
Also called Embeddings model, Text embedding model
An AI model that turns text, images or other content into embeddings: lists of numbers that capture meaning, used for search and comparison.
What is an embedding model?
An embedding model is a model whose only job is to turn content into an embedding: a long list of numbers that represents what the content means. Two passages about the same idea get similar numbers, even if they use different words. It doesn't write answers; it produces numbers for other systems to compare.
How does it work?
You send the model a piece of text (or, for multimodal embedding models, an image) and get back a vector, often hundreds or thousands of numbers long. In a typical RAG setup:
- Every chunk of your documents is embedded once and stored in a vector database.
- Each question is embedded with the same model.
- The database finds the stored vectors closest to the question's, often measured with cosine similarity, and returns those passages.
What is it commonly confused with?
- An embedding. The embedding is the output; the embedding model is what produces it.
- A chat model. Embedding models are much smaller and cheaper per request, and they can't generate text.
- Interchangeable models. Vectors from different embedding models can't be compared. If you switch models, you have to re-embed everything.
Why does it matter?
The embedding model decides how well a search system understands meaning, so it shapes the quality of semantic search and RAG. When choosing one, compare quality on content like yours (public leaderboards such as MTEB help), language support, maximum input length, vector size, which affects storage cost, and price per token.
