Modelembedding-modelsv1.0

BGE-M3

by BAAI · open-source · Last verified 2026-04-24

BGE-M3 from BAAI is a versatile open-source embedding model supporting dense, sparse, and multi-vector retrieval in a single unified model. It handles over 100 languages and long documents up to 8192 tokens, making it a top choice for multilingual and hybrid retrieval tasks. Available on Hugging Face for local and cloud deployment.

https://huggingface.co/BAAI/bge-m3 ↗

C—Below Average

Adoption: C+Quality: B+Freshness: ACitations: CEngagement: F

Specifications

License: Open Source
Pricing: open-source
Capabilities
Integrations
Use Cases
API Available: No
Parameters: 568M
Context Window: 8K tokens
Modalities: text
Training Cutoff: Late 2023
Tags: embeddings, open-source, multilingual, hybrid-retrieval, baai, dense-sparse
Added: 2026-04-24
Completeness: 60%

Index Score

Adoption

Quality

Freshness

Citations

Engagement

Need help choosing the right model?

Get Expert Guidance

Explore the full AI ecosystem on Agents as a Service