# Embedding Dimension

> Embedding dimension is the length of an embedding vector — how many numbers represent each text — trading capacity against storage and search cost.

**Embedding dimension is the length of the vector an [embedding](/glossary/embedding) model produces — 384, 768, 1536, 3072 numbers per text — setting the trade between how much meaning a vector can carry and what every vector costs to store and search.**

The economics are unforgiving because they're multiplicative: dimension × corpus size × bytes-per-float is your index's memory footprint, and search compute scales with it too. Double the dimensions and a 100M-vector index doubles in RAM — which is why dimension choice belongs in [vector-database](/glossary/vector-database) capacity planning, alongside [quantization](/glossary/quantization) of the vectors themselves.

Two modern developments take the sting out. **Matryoshka-style models** front-load information so vectors truncate gracefully — one model, several deployable sizes via an API parameter. And benchmark reality: today's well-trained 512–1,024-dim models frequently match yesterday's larger vectors, so the right process is empirical — test retrieval quality at two or three dimension settings on *your* corpus ([the embedding-selection guide](/guides/concepts/choosing-embeddings-2026)) and buy only the dimensions that earn their keep. One hard rule survives every choice: dimension is fixed per index — changing it means re-embedding everything.

---

_Source: https://agentscamp.com/glossary/embedding-dimension — Term on AgentsCamp._