ModelRefs / Embedding Model — AI Glossary
Embedding Model — AI Glossary
A model that converts text (or images) into dense fixed-dimensional vectors representing semantic meaning for similarity search.
Overview
Embedding models (OpenAI text-embedding-3-large, Cohere Embed v3, E5-large, GTE, Nomic Embed) encode text into 768–3072 dimensional vectors. Evaluated on MTEB (Massive Text Embedding Benchmark). Used for dense retrieval, semantic search, clustering, and deduplication. Dedicated embedding models dramatically outperform extracting embeddings from generative LLMs.
Reference details
| Topic | rag |
|---|---|
| Also known as | text embedding model, encoder model |
| Last reviewed | 2026-06-24 |
Related terms
Example: Dimensions are a storage decision
One million chunks embedded at 1,536 dimensions in float32 is 1,000,000 × 1,536 × 4 bytes ≈ 6.1 GB of raw vectors, before the index structure on top. At 768 dimensions it is about 3.1 GB; quantised to int8 it is roughly 1.5 GB at 1,536 dims. This is why Matryoshka-style models, trained so the leading dimensions stay useful when truncated, matter operationally: you can cut the vector in half and keep most of the retrieval quality — and you can measure exactly how much you lost on your own queries.
Commonly confused with
An embedding model is a bi-encoder: query and document are encoded separately, so document vectors are computed once and reused. A reranker is a cross-encoder that reads the pair together and cannot be precomputed. That is the whole reason a pipeline uses both — the cheap one narrows, the expensive one orders.
When to use it
Reach for it when:
- Retrieval, clustering, deduplication and classification over your own corpus
- Where document vectors can be computed ahead of time and cached
- Multilingual or domain corpora — choose on your own evaluation set, not a leaderboard
Reach for something else when:
- Extracting embeddings from a generative LLM's hidden states instead; purpose-trained encoders are better and cheaper
- Fine-grained ordering of a small candidate set, where a cross-encoder is the right tool
- Mixing vectors from two different models or dimensions in one index — they are not comparable
Where this appears on ModelRefs
Continue your research
Use these connected ModelRefs sections to compare alternatives, inspect implementation paths, and review the evidence and governance boundaries relevant to Embedding Model — AI Glossary.
Frequently asked questions
What is Embedding Model?
A model that converts text (or images) into dense fixed-dimensional vectors representing semantic meaning for similarity search.
Is Embedding Model the same as text embedding model?
Yes — text embedding model, encoder model are common aliases for Embedding Model.
What concepts are related to Embedding Model?
Closely related concepts include bi encoder, dense retrieval, embedding, reranking.