bge-small-en-v1.5

Small English BGE embedding checkpoint for low-overhead semantic search and retrieval on local or cost-sensitive deployments.

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

Lightweight English embedding pipelines where CPU or modest GPU serving matters more than chasing the largest model.

Who should use it

Lightweight English embedding pipelines where CPU or modest GPU serving matters more than chasing the largest model.
Builders who want local or self-hosted testing options.
RAG builders comparing retrieval quality, latency, and multilingual coverage.

Common workflows

Lightweight local embedding workflows
embedding workflows
rag workflows
retrieval workflows
semantic search workflows

Deployment and hardware notes

At 33.4M parameters (~0.13 GB in fp32) it embeds comfortably on CPU; a GPU mainly helps bulk-indexing throughput, not single-query latency.

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.
Lightweight English embedding pipelines where CPU or modest GPU serving matters more than chasing the largest model.
Prepend the query instruction 'Represent this sentence for searching relevant passages:' to short search queries; passages need no prefix in the v1.5 release.

Limitations

English only, and the 512-token window means long documents must be chunked before indexing. The 384-dimension vectors trade some ranking quality for the smallest index footprint in the BGE English family.
At 33.4M parameters (~0.13 GB in fp32) it embeds comfortably on CPU; a GPU mainly helps bulk-indexing throughput, not single-query latency.
Context window and limits: 512 tokens.
Verify the exact model card, provider docs, license, and serving support before production use.

Local workflow notes

Prepend the query instruction 'Represent this sentence for searching relevant passages:' to short search queries; passages need no prefix in the v1.5 release.

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

Platforms: Windows, macOS, Linux

Embedding spec

Memory~0.13 GB in fp32 (33.4M parameters) · ~0.07 GB in fp16Output384-dimension vectorsMax input512 tokens

Outputs 384-dimension vectors from CLS pooling. Pick it when index size and CPU serving dominate; step up to bge-base-en-v1.5 or bge-large-en-v1.5 when ranking quality on harder corpora matters more than footprint.