BGE-M3
Available nowHelp customers and teams find information across languages, even when they use different words.
What it does
BGE-M3, developed by BAAI, represents the meaning of text across more than 100 languages. It can connect a customer’s everyday question with a relevant product description or help article, rather than relying on an exact keyword match.
The original model supports dense, sparse and multi-vector retrieval. embeddings.app exposes its dense text vectors: one representation per input, suitable for a conventional vector search index. Its longer source context makes it useful for richer passages than the older English BGE models.
Language coverage
BAAI documents coverage of more than 100 languages. Individual language support and retrieval quality should be evaluated on your own queries.
Official language coverage ↗Where it helps
Global support
Connect questions in different languages with relevant help-center passages.
Product discovery
Match descriptions of a need with products whose wording is different.
Company knowledge
Find related policies, notes and documents across a multilingual collection.
When to choose it
A useful starting point for a multilingual service or a first search project. It is the lowest-priced available model here, so you can evaluate real customer queries with a small prepaid budget.
Use relevant passages rather than entire libraries as individual inputs. Sparse outputs and multi-vector scoring from the original model are not exposed by this service.
Official sources
Paraphrased from the provider’s documentation and original model information. Selection advice is our editorial guidance.
BAAI · BGE-M3 original model card ↗Reviewed October 3, 2026