BGE Reranker v2 M3
BGE Reranker v2-M3 is a 568M-parameter cross-encoder built on XLM-RoBERTa-large by BAAI, designed for second-stage reranking with an 8192-token context over query-document pairs in 100+ languages. Apache-2.0 licensed and shipped with text-embeddings-inference support, it is the most-downloaded reranker on the Hub and the multilingual baseline for production RAG.
Our verdict
The default open reranker for any production RAG in 2026. It is Apache-2.0 with 100-language coverage and 8K context. Pair it with bge-m3 or arctic-embed-l-v2 for a complete retrieve-rerank stack.
Overview
BGE Reranker v2-M3 is a 568M-parameter cross-encoder built on XLM-RoBERTa-large by BAAI, designed for second-stage reranking with an 8192-token context over query-document pairs in 100+ languages. Apache-2.0 licensed and shipped with text-embeddings-inference support, it is the most-downloaded reranker on the Hub and the multilingual baseline for production RAG.
Family & lineage
How this model relates to others in its lineage. Family members share architecture and training-data roots; parent / children edges record direct distillation or fine-tune relationships.
Strengths
- Multilingual 100+ languages with strong cross-lingual reranking (MIRACL average 72.84 at 1,024-token inference, reranking bge-m3's top 100, per the BAAI model card, https://huggingface.co/BAAI/bge-reranker-v2-m3)
- 8192-token cross-encoder context — full passages rerankable without truncation
- Apache-2.0, supported natively in TEI, vLLM, sentence-transformers, FastChat
- Best license/coverage/quality combination among open rerankers in production
Weaknesses
- 568M cross-encoder is heavy, so batch carefully
- FP16 speeds up computation with a slight quality drop, per the model card
- Larger bge-reranker-v2-gemma exists with better English quality if you can pay the 2B param cost
Quantization variants
Each quantization trades model quality for file size and VRAM. Q4_K_M is the most popular starting point.
| Quantization | File size | VRAM required |
|---|---|---|
| FP16 | 1.1 GB | 2 GB |
Get the model
HuggingFace
Original weights
Source repository — direct quantization required.
Hardware that runs this
Cards with enough VRAM for at least one quantization of BGE Reranker v2 M3.
Models worth comparing
Same parameter band, plus what's one tier above and below — so you can decide what actually fits your hardware.
Frequently asked
What's the minimum VRAM to run BGE Reranker v2 M3?
Can I use BGE Reranker v2 M3 commercially?
What's the context length of BGE Reranker v2 M3?
Source: huggingface.co/BAAI/bge-reranker-v2-m3
Reviewed by RunLocalAI Editorial. See our editorial policy for how we research and verify model claims.
Related — keep moving
Verify BGE Reranker v2 M3 runs on your specific hardware before committing money.