Model · BAAI · Embeddings
BGE-M3
BAAI's multi-functional embedding model. Supports dense, sparse, and multi-vector retrieval in one model — the strongest open-weights embedding option.
- Modality
- Embeddings
- License
- MIT (Open)
- Parameter size
- 568M
- Context window
- 8,192 tokens
- Released
- January 30, 2024
- Last verified
- September 8, 2026
- Runs locally
- Yes
What you need to run it
- Mac
- Yes. Any Apple Silicon Mac (M1 or newer) with at least 8 GB (16 GB recommended) of memory. Newer chips are faster, but memory is what decides whether it runs.
- Windows / Linux PC
- NVIDIA works best. A graphics card with at least 4 GB (8 GB recommended) of video memory.
- No graphics card
- Works. Runs on a normal computer's processor, just slower.
Small enough to run on a laptop CPU; Mac users can use the MLX builds or sentence-transformers.
Numbers are for: full 568M model, 16-bit (4-bit MLX build also exists). Our estimate from the model’s size — the maker publishes no figure.
Strengths
- MIT license, fully self-hostable
- 100+ language support
- Three retrieval modes (dense, sparse, multi-vector) from one model
Weaknesses
- Heavier than smaller embedding models — needs a GPU for high throughput
- Less polished than commercial options at the highest quality bar
Try it
| Where | Type | Notes |
|---|---|---|
| Hugging Face | weights | MIT |
| FlagEmbedding | local | Reference implementation |
Used in solutions
Official sources
- Model card model
- Paper paper
Change log
- — Added hardware requirements (Mac support, memory, NVIDIA). Source: estimate from size and community builds (model card HF BAAI/bge-m3).
- — Verified current. No bge-m4 exists; BGE-M3's successors are task-specific siblings rather than a new generation.
- — Initial entry.
Esc