Embedding Model
Also known as: Encoder model, Text encoder
The model that produces embeddings, typically a smaller encoder-only transformer.
A neural network trained to map text (or other modalities) into a fixed-size vector space where semantic similarity is preserved as geometric closeness. Examples: OpenAI text-embedding-3, BGE, E5, Cohere embed.
In practice
Choice of embedding model is the single biggest lever on RAG retrieval quality. Interviews probe BGE vs OpenAI vs domain-tuned models.
How it compares
An embedding model is the producer; embeddings are the output vectors.
Related topics
Related terms
Embeddings
Dense numeric vectors that capture meaning; close vectors = similar text.
Cosine Similarity
Compare two vectors by the cosine of their angle; the default similarity metric for embedding search.
Retriever
The fast first-stage search in RAG: bi-encoder or BM25 fetching a shortlist for the reranker or LLM.
Semantic Chunking
Split at topic boundaries by detecting big embedding drops between adjacent sentences.