Reference · Cheatsheet
Embedding dimensions cheat sheet
Last updated
Vectors have a **fixed length (dimensions)** per model — index and queries must match.
Vectors have a **fixed length (dimensions)** per model — index and queries must match.
Common models (OpenAI)
| Model | Dimensions | Notes |
|-------|------------|-------|
| `text-embedding-3-small` | 1536 (default) | Cost-effective RAG default |
| `text-embedding-3-large` | 3072 (default) | Higher quality, more storage |
| `text-embedding-ada-002` | 1536 | Legacy; still in old indexes |
Rules
- Same **model + dimensions** for index and query time
- Store **raw text** beside each vector
- Batch embed jobs — respect token limits per request
- Re-embed entire index when switching models
- Smaller dims ≠ always worse — test retrieval on your docs
Storage rough math
chunks × dimensions × 4 bytes (float32) ≈ index size
10,000 × 1536 × 4 ≈ 61 MB vectors only