Embedding
Also known as: vector embedding, text embedding, embeddings
A high-dimensional numerical vector representation of text, image, audio, or other content. Embeddings encode semantic meaning such that content with similar meaning sits close together in the vector space, enabling semantic search, retrieval-augmented generation (RAG), classification, and similarity scoring. The foundation primitive of every RAG-based agent.
Embedding model selection is under-discussed in 2026 enterprise agent procurement. Most programmes default to the foundation-model vendor's embedding model (text-embedding-3-large, voyage-3, gemini-embedding) without evaluating whether a smaller open-weight model (e5, gte, bge) would deliver comparable retrieval quality at materially lower cost. The right test is: rebuild the retrieval index against the candidate embedding model, run the deployment's eval set, compare retrieval@k and downstream answer quality. Most programmes skip this step and pay 5-10x more on embedding cost than the workload requires.
Articles that analyse this term
Primary sources
- OpenAI. Embeddings (API documentation)
- Anthropic. Embeddings — partner integrations