An embedding is a dense vector of numbers, produced by a machine learning model, that captures the meaning of a piece of data in a form that machines can compare mathematically. Feed text, an image, or audio into an embedding model and it returns a fixed-length array that encodes the semantic content of that input.
What makes embeddings powerful is that the model is trained so similar inputs produce nearby vectors. Two sentences with the same meaning land close together in the embedding space even if they share no words, while unrelated sentences land far apart. This turns the fuzzy notion of meaning into concrete geometry that can be searched with distance calculations.
Embeddings are the fundamental unit of every vector database. The quality of the embedding model determines the quality of everything built on top — search, recommendations, RAG, classification. Choosing a model means weighing its dimensionality, the domain it was trained on, its context length, language support, and benchmark scores, since these directly shape retrieval quality.