Embedding Dimension
Also called Vector Dimensionality
How many numbers each embedding uses — more can mean richer meaning but heavier storage.
Think of it like
Describing a wine with 5 traits vs. 50: more traits capture nuance but take longer to record and compare.
Example
Switching from a 1536-dim to a 768-dim embedding model halves your vector storage and speeds search, at some quality cost.
How it actually works
Each embedding is a fixed-length list of floats; the length is its dimension. Higher dimensions can encode finer distinctions but cost more memory, slower comparisons, and can hit diminishing returns. Some newer models support "Matryoshka" truncation, letting you shorten vectors for cheaper search when full fidelity is not needed.
For product teams
A direct lever on storage bill and latency; bigger is not automatically better.
For engineers
Fixed float-vector length; scales memory and distance-compute linearly, with Matryoshka models allowing lossy truncation.
Related
- Embeddings — A property of embeddings.
- Vector database — Affects storage and search in the vector database.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome