BGE-M3

BAAI

Unique tri-function embedding model: dense retrieval, sparse BM25-style retrieval, and multi-vector ColBERT-style retrieval in one model. 100+ languages, 8K context.

Embedding Local BGE Family vM3
Documentation Back to Models
Context Window
-
tokens
Parameters
568M
params
Source
Local
local
License
MIT
license

BGE Family 2 models

The full BGE line by generation — pricing and capabilities vary across the family.

BGE
BGE-M3 Current
Multilingual retrieval, hybrid search, RAG
Feb 2024

Local Model Specs

Architecture
XLM-RoBERTa Extended
Runtime
sentence-transformers / FlagEmbedding
Disk Size
2.2 GB

Details

Release Date
February 1, 2024
Knowledge Cutoff
-
Source
Local
License
MIT
Model ID
bge-m3
Last updated: March 13, 2026