Embedding
A list of numbers (a vector) that represents the meaning of a piece of text, so that similar meanings sit close together. The building block of semantic search and RAG.
What it is
An embedding model turns text into a fixed-length vector - commonly 384 to 3,072 numbers. Text with related meaning produces vectors that are near each other by a distance measure, usually cosine similarity.
Embedding APIs are priced per token of input, like an LLM, but far cheaper - often $0.02-0.15 per 1M.
Why it matters
Embeddings are what let you search a knowledge base by meaning instead of keywords. The cost has two parts: a one-time (or on-update) embedding pass over your documents, and ongoing embedding of each query.
Cost & infrastructure impact
Embedding a 10M-token corpus once at $0.10/1M is $1. The recurring cost is dominated by the vector database that stores and searches the results, not by the embedding calls.