Foundations

Embedding

A numerical vector that captures the semantic meaning of a piece of content.

An embedding is the output of a neural network that maps text, images, or audio into a fixed-length vector. Similar inputs produce vectors that are close together in the embedding space — 'invoice' and 'bill' end up near each other; 'invoice' and 'lasagna' do not.

Embedding models (text-embedding-3, Cohere embed, BGE, E5) are the workhorse behind semantic search, RAG, clustering, and classification. They are 100–1000x cheaper than LLM calls, so it is common to embed millions of documents up front.

Choice of embedding model matters more than most teams realize. Domain-specific models (legal, biomedical, code) often beat general-purpose ones for their niche.

Key points

Typical dimensions: 384, 768, 1024, 1536, 3072
Higher dimensions ≠ better — often the opposite for smaller corpora
Multilingual models cost 5–10% more but eliminate translation preprocessing
Re-embedding on model upgrade is a real operational cost

Common use cases

Semantic search

RAG document retrieval

Clustering and topic modeling

Zero-shot classification

Frequently asked

Related terms

Vector Database

A database optimized for storing and searching high-dimensional embeddings by similarity.

Retrieval-Augmented Generation (RAG)

A pattern that grounds LLM answers in your private documents by retrieving relevant snippets at query time.

Applying Embedding in your business?

We help enterprises design, ship, and govern AI systems end-to-end.

Start AI Readiness Check Talk to an expert

Back to glossary