Back to Home
Embeddings & Search

Embedding Dimensions

Embedding dimensions are the number of values used to represent text as a vector.

This is a simple learning demo. Real AI embeddings can have hundreds or thousands of dimensions.

Interactive Playground
Statistics
cat
Text
8
Dimensions
8
Vector Values
32 B
Est. Memory
Low
Search Cost
Dimensions as Features
📍
3D, like a map coordinate
(x, y, z), enough for position, not for meaning
🧩
128D, small model
Captures basic semantics; fast to store and search
🧠
1536D, large model (e.g. OpenAI)
Rich nuanced meaning; uses more memory and compute

💡 More dimensions = more detail, but also more storage. Choose based on your accuracy vs speed trade-off.

Vector Visualization
Vector Values 8 dimensions
Memory Usage (relative) 32 bytes
2 dims (8 B) 128 dims (512 B)
Size Comparison
2 Dimensions
Small representation
128 Dimensions
Larger representation
Real Model Dimensions
ModelDims
OpenAI
text-embedding-3-small 1,536
text-embedding-3-large 3,072
Google
text-embedding-004 768
Sentence Transformers
all-MiniLM-L6-v2 384
all-mpnet-base-v2 768
Alibaba / Qwen
Qwen3-Embedding-0.6B 1,024
Qwen3-Embedding-8B 4,096
You don't choose what each dimension encodes, the model learns that during training. Dimension #42 might capture "formality", but there's no readable label.
More dimensions = more storage and slower search. A 1536-D vector takes 12× more space than a 128-D one. Choose based on your accuracy vs speed trade-off.
💡
Key Takeaway

More dimensions can store more detail, but they also use more memory and computation.