Model family
BGE Models
BGE models cover embeddings, reranking, retrieval, semantic search, and vector database workflows for RAG builders.
Best for
Embedding
Use this family hub to compare BGE variants for embedding workflows, then open the detail page for deeper deployment notes.
Reranking
Use this family hub to compare BGE variants for reranking workflows, then open the detail page for deeper deployment notes.
RAG
Use this family hub to compare BGE variants for rag workflows, then open the detail page for deeper deployment notes.
Search
Use this family hub to compare BGE variants for search workflows, then open the detail page for deeper deployment notes.
Source box
This family currently includes 9 records tied to an exact published checkpoint. Identity is recorded per model so a representative checkpoint is never treated as the whole family.
Identity checked: 2026-08-21
Artifact identity does not establish licence or context truth. Those checks remain separate.
Variants
BGE models grouped by workflow
Embedding and reranking
BGE Reranker v2 M3
BAAI · BGE
Best for: RAG builders who need a practical reranker after Qdrant, Chroma, pgvector, or other vector search.
Local: Commonly used in local RAG stacks as a reranking step after vector search.
bge-m3
BAAI · BGE
Best for: RAG teams that want one embedding model for multilingual search, long documents, and retrieval experiments beyond plain dense vectors.
Local: No query instruction needed — unlike the BGE v1.5 line, queries and passages are embedded bare, which simplifies pipelines.
bge-large-en-v1.5
BAAI · BGE
Best for: English retrieval stacks that want a stronger embedding baseline for search and RAG before moving to multilingual models.
Local: Short queries benefit from the 'Represent this sentence for searching relevant passages:' instruction; the v1.5 release made it optional rather than required.
bge-base-en-v1.5
BAAI · BGE
Best for: Teams building English search or RAG systems on modest GPUs without dropping to the smallest embedding tier.
Local: Use the v1.5 query instruction for short queries ('Represent this sentence for searching relevant passages:'); documents are embedded without it.
bge-small-en-v1.5
BAAI · BGE
Best for: Lightweight English embedding pipelines where CPU or modest GPU serving matters more than chasing the largest model.
Local: Prepend the query instruction 'Represent this sentence for searching relevant passages:' to short search queries; passages need no prefix in the v1.5 release.
bge-large-zh-v1.5
BAAI · BGE
Best for: Chinese embedding and retrieval workflows
Local: Chinese queries use the model card's own instruction prefix (为这个句子生成表示以用于检索相关文章：); passages are embedded bare.
bge-reranker-large
BAAI · BGE
Best for: Teams that already retrieve a candidate set and want a stronger final ranking step before passing context to an LLM.
Local: A drop-in upgrade for bge-reranker-base pipelines: same 512-token pair format, same score-only output, roughly double the compute per pair.
bge-reranker-base
BAAI · BGE
Best for: Builders who want a practical reranking step in RAG or search without adding a heavyweight cross-encoder to every request.
Local: Feed it (query, passage) pairs from your first-stage retriever; it returns a relevance score per pair and produces no embeddings to store.
bge-embedding-gemma2
BAAI · BGE
Best for: Gemma-based embedding experiments
Local: Budget throughput carefully — bulk-indexing a large corpus through a 9.2B model is orders of magnitude more compute than this page's sub-1B siblings.
Compare
All BGE models in the directory
| Model | Type | Best for | Local runner notes | License | Detail |
|---|---|---|---|---|---|
| BGE Reranker v2 M3 | Reranking | RAG builders who need a practical reranker after Qdrant, Chroma, pgvector, or other vector search. | Commonly used in local RAG stacks as a reranking step after vector search. | Apache 2.0 | Open |
| bge-m3 | Embedding | RAG teams that want one embedding model for multilingual search, long documents, and retrieval experiments beyond plain dense vectors. | No query instruction needed — unlike the BGE v1.5 line, queries and passages are embedded bare, which simplifies pipelines. | MIT | Open |
| bge-large-en-v1.5 | Embedding | English retrieval stacks that want a stronger embedding baseline for search and RAG before moving to multilingual models. | Short queries benefit from the 'Represent this sentence for searching relevant passages:' instruction; the v1.5 release made it optional rather than required. | MIT | Open |
| bge-base-en-v1.5 | Embedding | Teams building English search or RAG systems on modest GPUs without dropping to the smallest embedding tier. | Use the v1.5 query instruction for short queries ('Represent this sentence for searching relevant passages:'); documents are embedded without it. | MIT | Open |
| bge-small-en-v1.5 | Embedding | Lightweight English embedding pipelines where CPU or modest GPU serving matters more than chasing the largest model. | Prepend the query instruction 'Represent this sentence for searching relevant passages:' to short search queries; passages need no prefix in the v1.5 release. | MIT | Open |
| bge-large-zh-v1.5 | Embedding | Chinese embedding and retrieval workflows | Chinese queries use the model card's own instruction prefix (为这个句子生成表示以用于检索相关文章：); passages are embedded bare. | MIT | Open |
| bge-reranker-large | Reranking | Teams that already retrieve a candidate set and want a stronger final ranking step before passing context to an LLM. | A drop-in upgrade for bge-reranker-base pipelines: same 512-token pair format, same score-only output, roughly double the compute per pair. | MIT | Open |
| bge-reranker-base | Reranking | Builders who want a practical reranking step in RAG or search without adding a heavyweight cross-encoder to every request. | Feed it (query, passage) pairs from your first-stage retriever; it returns a relevance score per pair and produces no embeddings to store. | MIT | Open |
| bge-embedding-gemma2 | Embedding | Gemma-based embedding experiments | Budget throughput carefully — bulk-indexing a large corpus through a 9.2B model is orders of magnitude more compute than this page's sub-1B siblings. | Gemma Terms of Use | Open |