what embedding model should I use for multilingual search?

For multilingual search, the best embedding model depends on your priorities:

Good default choice

intfloat/multilingual-e5-large

• Strong multilingual retrieval performance
• Works well for semantic search across many languages
• Commonly used and easy to adopt
• Best when you want one solid model for query/document search

Other strong options

BAAI/bge-m3

• Very strong multilingual model
• Supports dense, sparse, and multi-vector retrieval
• Good if you want flexibility or better retrieval quality in more complex setups

intfloat/multilingual-e5-base

• Smaller/faster than large
• Good if you need lower latency or lower cost
• Slightly less accurate than the large version

If you need hosted/API models

Look at:

• OpenAI text-embedding-3-large or text-embedding-3-small
• These are strong and multilingual, but if you need local/open-source deployment, the models above are better fits.

Practical recommendation

• Best overall open-source default: intfloat/multilingual-e5-large
• Best if you want state-of-the-art flexibility: BAAI/bge-m3
• Best lightweight option: intfloat/multilingual-e5-base

Important tip

For multilingual search, results often improve a lot if you:

• use the same embedding model for both queries and documents
• normalize text consistently
• test with your actual languages and search data
• consider reranking after retrieval for better relevance

If you want, I can recommend a model based on:

1. your languages,
2. whether you need open-source vs API, and
3. your latency/cost constraints.