What should I use for multilingual embeddings across English and Spanish content?

For English + Spanish multilingual embeddings, the safest default is to use a multilingual embedding model rather than separate monolingual ones.

Good choices

• OpenAI text-embedding-3-large or text-embedding-3-small
  Strong multilingual coverage, easy to use, and works well for cross-lingual retrieval.
• Sentence Transformers multilingual models
  Good if you want to run embeddings locally:
  • paraphrase-multilingual-MiniLM-L12-v2
  • distiluse-base-multilingual-cased-v2
• Cohere multilingual embeddings
  Also strong for cross-language search/retrieval.

Recommendation

If you want the simplest and generally best option for mixed English/Spanish search, use:

• text-embedding-3-large for highest quality
• text-embedding-3-small if you want lower cost / faster latency

Why multilingual embeddings?

They map semantically similar English and Spanish text into the same vector space, so:

• Spanish queries can retrieve English documents
• English queries can retrieve Spanish documents
• You avoid building and maintaining two separate systems

Practical tips

• Keep all content in one embedding model
• Normalize text consistently
• If you do retrieval, test with:
  • English query → Spanish doc
  • Spanish query → English doc
  • same-language matches
• If your corpus is highly domain-specific, consider fine-tuning or adding a reranker

If you want, I can also recommend the best option based on your constraints:

• cloud vs local
• budget
• retrieval vs clustering vs semantic search
• expected corpus size