How to pick an embedding model

Embeddings power RAG retrieval — pick for **language, domain, and cost**, not buzzwords.

1. **Match language** — multilingual docs need multilingual embeddings

2. **Check dimension size** — higher ≠ always better; affects DB storage

3. **Run a retrieval test** — 20 questions, did the right chunk land in top-3?

4. **Compare latency & price** at your document volume

5. **Lock one model** per index — don't mix embeddings in one shelf

| Use case | Hint |

|----------|------|

| English general docs | Small sentence-transformer class models |

| Code repos | Code-aware embedding models |

| Multilingual support | Multilingual E5 / similar |

Swap models only if hit rate jumps on **your** data.

**Try the lessons:** `what-are-embeddings` · `vectors-and-similarity`