9.2B- Context
- 8Ktokens
- Released
- 2y ago
- Size
- 568Mparameters
BGE Reranker V2 M3 is an open-weights reranker from BAAI that belongs to its BGE reranker v2 series, which the model card recommends for multilingual retrieval and for Chinese or English scenarios where efficiency matters. Unlike an embedding model, it takes a query and passage as input and directly outputs a relevance score that a sigmoid function maps to a value between 0 and 1. It is built for reranking retrieved documents, such as reordering the top results from retrievers like bge-m3, and can be fine-tuned on custom query-passage data.
How it compares
Additional source evidence
- Source model
- baai/bge-reranker-v2-m3
- Source family
- bge
- Source type
- reranking
- Last updated
- 2024-03-15
- Model context
- 8K
- Model output limit
- —
- Open weights
- Yes
- File attachments
- No
- Reasoning
- No
- Tool calling
- No
- Temperature control
- No
Downloadable weights
Complete source record (JSON)
{
"id": "baai/bge-reranker-v2-m3",
"name": "BGE Reranker V2 M3",
"description": "Lightweight multilingual cross-encoder that scores query-document relevance",
"family": "bge",
"type": "reranking",
"last_updated": "2024-03-15",
"attachment": false,
"reasoning": false,
"tool_call": false,
"temperature": false,
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"limit": {
"context": 8192,
"output": 0
},
"release_date": "2024-03-15",
"open_weights": true,
"license": "Apache-2.0",
"weights": [
{
"label": "Hugging Face",
"url": "https://huggingface.co/BAAI/bge-reranker-v2-m3"
}
]
}Frequently asked questions
Running locally
By our estimate, BGE Reranker V2 M3 needs about 0.9 GB at Q4, 1.2 GB at Q8 and 1.8 GB at FP16 of VRAM. That covers the weights plus headroom for a short context, and longer contexts need more.
Yes, by our estimates BGE Reranker V2 M3 fits on an 8 GB GPU at FP16 (about 1.8 GB) and at higher precision on larger cards.
Yes, a Mac with at least 8 GB of unified memory should run BGE Reranker V2 M3 at Q4, which needs an estimated 0.9 GB. This assumes macOS gives the GPU about 75% of unified memory.
Privacy and data
Yes, BGE Reranker V2 M3 has open weights, so you can run it on your own hardware and keep prompts on your own machines.
Capabilities and specs
BGE Reranker V2 M3 has a 8K token context window.
No, BGE Reranker V2 M3 takes text input only.
No, BGE Reranker V2 M3 is not listed with tool calling support.
No, BGE Reranker V2 M3 is not listed with a reasoning mode.
Quality and benchmarks
Similar models from other labs include OLMo 1.7 7B (Ai2), OpenELM 1.1B (Apple) and OpenELM 270M (Apple). They are the closest matches by kind and release date.
OLMo 1.7 7BBGE Reranker V2 M3 vs OLMo 1.7 7BOpenELM 1.1BBGE Reranker V2 M3 vs OpenELM 1.1BOpenELM 270MBGE Reranker V2 M3 vs OpenELM 270M
Licensing and access
BGE Reranker V2 M3 has open weights under the Apache-2.0 license, so you can download and run it yourself.
Yes, BGE Reranker V2 M3 is released under the Apache 2.0 license, which allows commercial use. Check the license text for attribution terms.
The BGE Reranker V2 M3 weights are on Hugging Face at BAAI/bge-reranker-v2-m3.
Basics
BGE Reranker V2 M3 was made by BAAI. It was released on March 15, 2024.
Yes, BGE Reranker V2 M3 has open weights, so you can run it on your own hardware without API fees.
Generated from Vertical Tabs data, updated November 2, 2026.
More in Bge
9.2B567M
Discussion
No comments yet
Be the first to share your thoughts.