bge-base-en-v1.5

Mid-sized English BGE embedding checkpoint for semantic search and RAG when you want a balance between retrieval quality and serving cost.

BAAI · BGE

Editorial review

Reviewed byOpenSourcesAI EditorialLast updatedAugust 2026SourcesExact model card

Model checkpoints, context windows, provider support, local runtime compatibility, and license terms can change quickly. Verify the exact model card before production or commercial use.

Best for

Teams building English search or RAG systems on modest GPUs without dropping to the smallest embedding tier.

Who should use it

Teams building English search or RAG systems on modest GPUs without dropping to the smallest embedding tier.
Builders who want local or self-hosted testing options.
RAG builders comparing retrieval quality, latency, and multilingual coverage.

Common workflows

Balanced English embedding workflows
embedding workflows
rag workflows
retrieval workflows
semantic search workflows

Deployment and hardware notes

109.5M parameters (~0.44 GB in fp32) still serves on CPU at moderate query rates; a modest GPU covers bulk re-indexing.

License and usage notes

MIT. Open weights. Verify the exact model card and license terms for the checkpoint or hosted provider you use.

Strengths

Open weights model option for BGE workflows.
Teams building English search or RAG systems on modest GPUs without dropping to the smallest embedding tier.
Use the v1.5 query instruction for short queries ('Represent this sentence for searching relevant passages:'); documents are embedded without it.

Limitations

English only with a 512-token input ceiling, so retrieval quality depends on sensible chunking. Its 768-dimension vectors double the index size of bge-small's 384.
109.5M parameters (~0.44 GB in fp32) still serves on CPU at moderate query rates; a modest GPU covers bulk re-indexing.
Context window and limits: 512 tokens.
Verify the exact model card, provider docs, license, and serving support before production use.

Local workflow notes

Use the v1.5 query instruction for short queries ('Represent this sentence for searching relevant passages:'); documents are embedded without it.

Local runtimes: Ollama where supported, LM Studio where supported, llama.cpp where supported, Transformers

Platforms: Windows, macOS, Linux

Embedding spec

Memory~0.44 GB in fp32 (110M parameters) · ~0.22 GB in fp16Output768-dimension vectorsMax input512 tokens

The middle size of the English BGE v1.5 trio: measurably stronger retrieval than bge-small at roughly a third of bge-large's compute. 768-dimension CLS-pooled vectors.