Embedding dimension
Embedding dimension is the number of values in each vector an embedding model outputs. More dimensions can hold more detail, but every extra value adds storage, memory and search cost per item.
Also called: dimensions, vector dimension, Matryoshka
By Divya Sharma · Checked Nov 15, 2026 · How we review
When you pick an embedding model you also pick a vector length. OpenAI's text-embedding-3-small returns 1536 numbers by default and text-embedding-3-large returns 3072. Every document chunk you store carries that many numbers, so the dimension multiplies straight into the size of your vector database and the work done on each search.
Does it matter?
Yes, mostly for cost and speed. A larger vector does not guarantee better retrieval: model training matters more than raw length. A smaller vector from a strong model can beat a larger one from a weaker model. Check retrieval quality on your own data rather than assuming bigger is better.
Shortening vectors
Some newer models are trained so you can cut a vector down and keep most of its usefulness. OpenAI lets you pass a `dimensions` parameter that drops numbers from the end of the vector. This works because of Matryoshka embeddings, a training method that packs the most important information into the first values.
Practical notes
The dimension is fixed when you create an index, so changing it means re-embedding and re-indexing. Query and document vectors must have the same length. Storage per vector also depends on number format: quantization of vectors to fewer bits is a separate way to save space. The embedding models guide lists dimensions for the models we review.
Related terms
- Vector database
- A vector database stores embeddings alongside their source records and finds the stored vectors closest to a query vector, usually with an approximate index so search stays fast across millions of items.
- Embedding
- An embedding is a list of numbers (a vector) that represents a piece of text, an image or other data, produced so that items with similar meaning end up with vectors close together.
- Matryoshka embeddings
- Matryoshka embeddings are vectors trained so that their first values carry the most information. You can cut them to a shorter length and still get useful search results, which saves storage and speeds up search.
- Quantization
- Quantization is a technique that stores a model's weights, and sometimes activations, with fewer bits, such as 8 or 4 instead of 16, so the model needs less memory and often runs faster, at some cost in accuracy.
- Approximate nearest neighbor search
- Approximate nearest neighbor (ANN) search finds vectors close to a query without checking every stored vector. It gives up a guarantee of the exact best match in exchange for much faster search on large collections.