Which embedding model should I use with Atlas Vector Search?

For general text (help docs, product descriptions, articles): OpenAI text-embedding-3-small at 1536 dimensions, or Voyage AI voyage-3-lite at 512 dimensions for a cheaper cost tier. Both give strong quality. For domain-specific embeddings (legal, medical, code): Voyage AI voyage-code-3 or a fine-tuned model. For multi-lingual: Cohere embed-multilingual-v3. We recommend text-embedding-3-small as the default and swap only when your evaluation suite shows a real quality difference on your actual queries.