BGE-M3

What is BGE-M3?

BGE-M3 is BAAI's open-weights language model. Released in 2024-01, it ships in 1 size (567M) and is licensed MIT. The most popular use case is multilingual embeddings (100+ languages).

VRAM and hardware

The 567M size fits in 1 GB of VRAM.

Size	Min VRAM (Q4_K_M)	Recommended VRAM	Tokens/sec on 3090
567M	1 GB	2 GB	~34 tok/s

How to run BGE-M3 locally

Option 1: Ollama (simplest)

ollama pull bge-m3
ollama run bge-m3

Option 2: Mullama (production)

mullama pull bge-m3
mullama run bge-m3

Option 3: llama.cpp (CLI)

# Download a GGUF from Hugging Face (search "bge-m3 gguf")
./llama-cli -m bge-m3.Q4_K_M.gguf -p "Hello, AI!"

Option 4: Python with Mullama or llama-cpp-python

from mullama import Model, Context
model = Model.load("bge-m3.Q4_K_M.gguf", n_gpu_layers=99)
ctx = Context(model, n_ctx=4096)
print(ctx.generate("Hello, AI!", 256))

What you can build with BGE-M3

• multilingual RAG
• hybrid retrieval
• long-document

Hardware it fits on

• Apple Silicon (8GB): 567M in MLX
• Apple Silicon (16GB): 567M in MLX
• Apple Silicon (32GB): 567M in MLX or GGUF
• Apple Silicon (64GB+): 567M in MLX
• RTX 3090 (24GB): 567M at Q4_K_M
• RTX 4090 (24GB): same as 3090, ~30% faster
• RTX 5090 (32GB): 567M at Q8_0
• 2× RTX 6000 Ada (96GB total): 567M at Q4_K_M

Related models in the BGE family

This is the only BGE-family model in the catalog right now.

Other models you might consider

• Llama 3.1 (Meta) — 8B, general-purpose chat, agentic workflows, code
• Llama 3.2 (Meta) — 1B, low-end hardware, mobile, edge deployment
• Llama 3.2 Vision (Meta) — 11B, vision + chat on desktop GPUs
• Llama 3.3 (Meta) — 70B, best-in-class 70B for desktops
• Llama 4 (Meta) — Scout 17B (109B MoE), vision + long context (10M tokens)
• Qwen 2.5 (Alibaba) — 0.5B, strongest 7B-32B on consumer hardware