MODEL · BAAI · 568M (XLM-ROBERTA-LARGE BASE)
BGE-M3
BAAI's multi-functionality + multilingual (170+ languages) + multi-granularity embedding. The default "just use it" RAG embedding since early 2024. As of 2026 it is no longer the top-quality pick — `Qwen3-Embedding` (0.6B / 4B / 8B, Apache 2.0) now leads MTEB overall — but BGE-M3 remains the sharpest pick for cheap, broad multilingual breadth at 568M.
License: MIT · Context: 8192 tokens · Released: February 2024
The decision in five lines
- The call
- Consider — runnable locally, family reference
- Best for
- Local evaluation and family reference
- Runs on
- 23 hardware picks fit (cheapest: Intel Arc B580 12 GB · $249)
- Watch out
- Max-quality general/English retrieval — Qwen3-Embedding-8B (Apache 2.0) ranks #1 on MTEB and is the better default when you have the VRAM.
- Evidence
- Estimated
- 568M (XLM-RoBERTa-large base)
- PARAMETERS
- EMBEDDING
- TYPE
- 8192
- CONTEXT
- ~1–2 GB
- VRAM AT Q4
Where we recommend this
This model isn’t currently in an active planner slot. See the runner notes below if you’re running it anyway.
The call
BAAI's multi-functionality + multilingual (170+ languages) + multi-granularity embedding. The default "just use it" RAG embedding since early 2024. As of 2026 it is no longer the top-quality pick — `Qwen3-Embedding` (0.6B / 4B / 8B, Apache 2.0) now leads MTEB overall — but BGE-M3 remains the sharpest pick for cheap, broad multilingual breadth at 568M.
When not to use: Max-quality general/English retrieval — Qwen3-Embedding-8B (Apache 2.0) ranks #1 on MTEB and is the better default when you have the VRAM. Use BGE-M3 when you want the smallest model that still covers 170+ languages, or a CPU-friendly 568M retriever.
Runner notes
Ollama tag `bge-m3`. Also natively in FlagEmbedding, sentence-transformers, llama.cpp. 568M params run fine on CPU for small corpora. For top quality step up to `Qwen/Qwen3-Embedding-8B` (or the 0.6B/4B for lighter rigs).
Hardware that fits
Every hardware pick whose memory fits this model at the quant we recommend. Sorted cheapest-first — the top row is your best-value fit. Click through for the full buyer’s guide.
- Intel Arc B580 12 GBPerfect · 5.9× 12 GB · $249–$299
- NVIDIA RTX 3060 12 GBPerfect · 5.9× 12 GB · $280–$400
- Minisforum UM890 ProPerfect · 11.8× 32 GB DDR5 (shared) · $463–$580 all-in
- RTX 5060 Ti 16 GBPerfect · 7.9× 16 GB · $560–$610
- AMD Radeon RX 9070 XTPerfect · 7.9× 16 GB · $649–$779
- Mac Mini M4 16 GBPerfect · 5.3× 16 GB unified · $799 (new floor) / $499–$599 (eBay/residuals)
- AMD Radeon RX 7900 XTXPerfect · 11.8× 24 GB · $810 used / ~$1,340 new
- NVIDIA RTX 3090 (used, single)Perfect · 11.8× 24 GB · $950–$1,200
- NVIDIA RTX 5070 TiPerfect · 7.9× 16 GB · $980–$1,300
- NVIDIA RTX 5080Perfect · 7.9× 16 GB · $1,250–$1,400
- MacBook Air M5 24 GBPerfect · 7.9× 24 GB unified · $1,499–$1,899
- Mac Mini M4 Pro 24 GBPerfect · 7.9× 24 GB unified · $1,599
- Dual RTX 3090 (used)Perfect · 23.6× 48 GB · $1,800–$2,500 all-in
- NVIDIA RTX 4090Perfect · 11.8× 24 GB · $2,200–$2,800
- Framework Desktop (Ryzen AI Max+ 395)Perfect · 42.1× 128 GB unified · $2,459–$2,851 (128 GB config)
- M5 Pro MacBook Pro 48 GBPerfect · 15.8× 48 GB unified · $2,999–$3,599
- NVIDIA RTX 5090Perfect · 15.7× 32 GB · $3,500–$4,300
- NVIDIA RTX A6000 (48 GB, used)Perfect · 23.6× 48 GB ECC · $3,500–$4,500
- Mac Studio M4 Max 64 GBPerfect · 21.0× 64 GB unified · $3,799
- NVIDIA DGX SparkPerfect · 42.1× 128 GB unified · $4,699
- M5 Max MacBook Pro 64 GBPerfect · 21.0× 64 GB unified · ~$5,199 (est.; June 25 2026 increase)
- Mac Studio M3 Ultra 96 GBPerfect · 31.6× 96 GB unified · $5,299
- Dual RTX 5090Perfect · 31.4× 64 GB (2×32) · $8,500–$10,500
Next step
Find-by-model — see what hardware runs this→