bge-reranker

Open weights · BAAI · free to self-host · Updated 12 Oct 2026

The bge-reranker family from BAAI (Beijing Academy of Artificial Intelligence) are open-weight cross-encoders trained specifically for passage reranking. They're the standard self-hosted option: free to download, drop into sentence-transformers, and the open-weight baseline most other rerankers measure themselves against.

Model variants

Recommendation: start with bge-reranker-v2-m3. It's multilingual, small enough for CPU, and works fine for moderate traffic. Upgrade to the Gemma variant only if you have GPU and need maximum quality.

Benchmarks

BAAI evaluated the two generations on different benchmarks, so neither column is complete: BEIR (English retrieval) for v2, C-MTEB reranking (mostly Chinese) for v1.

* All figures are BAAI's own. BEIR: 15 datasets, reranking the top 100 results from bge-large-en-v1.5 (FlagEmbedding llm_reranker evaluation). C-MTEB: average MAP over six reranking sets — T2Reranking, its two cross-language variants, MMarcoReranking, CMedQAv1 and CMedQAv2 (FlagEmbedding reranker README). Other evaluators get different BEIR numbers from their own setups — Jina's paper has v2-m3 at 56.51* — so compare models within one source. Until October 2026 this table showed BEIR and MS MARCO figures for all four models that we could not trace to any source; they have been removed.

Quick start

Install

pip install sentence-transformers

Rerank a list of passages

from sentence_transformers import CrossEncoder

model = CrossEncoder("BAAI/bge-reranker-v2-m3", max_length=512)

query = "How do I add reranking to my RAG pipeline?"
passages = [
    "Rerankers score each query-passage pair with a cross-encoder.",
    "BM25 is a classical keyword-based retrieval method.",
    "London is the capital of the United Kingdom.",
    "Two-stage retrieval: retrieve 50 candidates, rerank to top 5.",
]

scores = model.predict([(query, p) for p in passages])
ranked = sorted(zip(scores, passages), reverse=True)

for score, text in ranked:
    print(f"{score:.4f}  {text[:80]}")

In a RAG pipeline

from sentence_transformers import CrossEncoder

reranker = CrossEncoder("BAAI/bge-reranker-v2-m3")

def rag_answer(query: str, vector_db, llm) -> str:
    # Stage 1: retrieve wide
    candidates = vector_db.search(query, top_k=50)
    # Stage 2: rerank tight
    scores = reranker.predict([(query, c) for c in candidates])
    top5 = [c for _, c in sorted(zip(scores, candidates), reverse=True)[:5]]
    # Stage 3: generate
    return llm.complete(f"Context:\n" + "\n\n".join(top5) + f"\n\nQ: {query}")

Pros and cons

Other models