BGE-reranker-v2-m3
BAAI · Reranker · 0.6B · 8k context · Released 1 February 2024
BGE-reranker-v2-m3 is a cross-encoder that takes a query and a candidate passage together and outputs a relevance score, used to re-order the shortlist from a first-stage retriever. It is multilingual, based on the same BGE-M3 family as the popular embedding model, and small enough to run on modest hardware or the CPU. It has been a default reranker in local RAG stacks for long enough to be well integrated across the major frameworks, which is much of its appeal. Newer rerankers such as Qwen3-Reranker score higher on several benchmarks, but BGE's maturity and broad tooling support keep it a reliable choice.
Strengths
- Mature and widely supported across RAG frameworks
- Multilingual and lightweight, running on modest hardware
- Apache 2.0, so no commercial-use conditions
Weaknesses
- Newer rerankers score higher on several reranking benchmarks
- Usage examples default to a short input length, which needs raising for long passages
- The comparison figures below were measured by a competitor, not by BAAI
Hardware requirements
| Quantisation | Approx. VRAM | Notes |
|---|---|---|
| FP16 | ~2GB | A 0.6B model is tiny, so full precision is the norm and fits anything |
Also runs on CPU (slower). Optimised builds available for Apple Silicon.
What you'd need to run this
Roughly what a machine to run this would need, at up to three levels of quality. Memory is the deciding factor.
Minimum to run it
FP16 · ~2GB needed
One 12GB GPU
NVIDIA GeForce RTX 3060 12GBor a Mac or mini-PC with unified memory, if you prefer no discrete GPU, Mac mini M5 Pro .
At least 32GB of system RAM alongside the card.
around £700–£1,100
What else 12GB runs →Licence
Apache 2.0 — read the licence
Benchmarks
| Benchmark | Score | Source | As of |
|---|---|---|---|
| MTEB-R | 57.03 | Qwen3-Reranker card comparison table (competitor-measured) | June 2025 |
| CMTEB-R | 72.16 | Qwen3-Reranker card comparison table (competitor-measured) | June 2025 |
How it compares
How this model’s reported scores sit against other models we cover, on the same benchmarks. This model is highlighted.
MTEB-R
higher is better- Qwen3-Reranker 0.6B 65.80
Qwen3-Reranker model card · June 2025
- BGE-reranker-v2-m3 57.03
Qwen3-Reranker card comparison table (competitor-measured) · June 2025
BGE-reranker-v2-m3: common questions
- What hardware do I need to run BGE-reranker-v2-m3?
- At its most compressed (FP16) it needs roughly 2GB of VRAM, and about 2GB for good quality. VRAM figures are approximate and depend on context length and settings.
- Is BGE-reranker-v2-m3 free for commercial use?
- Yes. BGE-reranker-v2-m3 is licensed under Apache 2.0, which permits commercial use with no meaningful conditions.
- Can I run BGE-reranker-v2-m3 on Apple Silicon?
- Yes. BGE-reranker-v2-m3 has builds optimised for Apple Silicon, through MLX or GGUF on a Mac.
- Does BGE-reranker-v2-m3 run on CPU?
- Yes, BGE-reranker-v2-m3 can run on the CPU, though generation is slower than on a GPU.
- What is BGE-reranker-v2-m3's context window?
- BGE-reranker-v2-m3 has a context window of 8,192 tokens, about 8k.
Availability
Recommended for
- A mature, well-supported reranker for a local RAG system
- Multilingual reranking on modest hardware
- Pairing with the BGE-M3 embedding model
Related models
Run it with
Related guides
Catalogue entry last verified 18 August 2026. Specifications change; verify anything you are about to spend money on.