Evidence-first authority page
What Is an Embedding Model (vs. a Chat Model)?
The distinction between an embedding model and a chat/completion model, and why you need both for a typical RAG system.
Direct answer
An embedding model converts text into a numerical vector representing its meaning, used for similarity search — it does not generate text. A chat or completion model generates text responses. A typical RAG system uses an embedding model for retrieval and a separate chat model to generate the final answer from retrieved context; conflating the two is a common source of confusion for teams new to building RAG systems.
Last reviewed 2026-08-29
This page remains noindex while its page-level source receipts are completed and verified. It is available for review, not presented as publication-ready authority.
Two different jobs in a RAG pipeline
The embedding model's only job is producing a vector that captures a piece of text's meaning, used to find similar content via distance calculations. The chat model's job is generating a coherent, relevant response given a prompt and retrieved context. These are typically different models, sometimes from different providers, chosen independently based on each one's specific performance for its specific job.
Why mismatched embedding models cause problems
Vectors from two different embedding models aren't comparable to each other — if you switch embedding models, you generally need to re-embed your entire document store, not just new documents, since old and new vectors won't be meaningfully comparable in the same similarity search.