e5-small-v2

Small E5 embedding checkpoint for lightweight English semantic search and retrieval on modest hardware.

Microsoft / intfloat · E5

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

Local or cost-sensitive embedding pipelines that need a practical English retrieval model without the overhead of larger E5 releases.

Who should use it

Local or cost-sensitive embedding pipelines that need a practical English retrieval model without the overhead of larger E5 releases.
Builders who want local or self-hosted testing options.
RAG builders comparing retrieval quality, latency, and multilingual coverage.

Common workflows

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

Deployment and hardware notes

~0.13 GB in fp32 (33.4M parameters) — genuinely CPU-class; it embeds and searches responsively with no GPU at all.

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 E5 workflows.
Local or cost-sensitive embedding pipelines that need a practical English retrieval model without the overhead of larger E5 releases.
Same contract as its siblings: 'query: ' and 'passage: ' prefixes on everything, mean pooling, cosine similarity.

Limitations

The fastest v2 E5 gives up ranking quality on harder corpora, stays English-only, and still requires the E5 prefixes at 512 tokens or less per input.
~0.13 GB in fp32 (33.4M parameters) — genuinely CPU-class; it embeds and searches responsively with no GPU at all.
Context window and limits: 512 tokens.
Verify the exact model card, provider docs, license, and serving support before production use.

Local workflow notes

Same contract as its siblings: 'query: ' and 'passage: ' prefixes on everything, mean pooling, cosine similarity.

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

384-dimension vectors keep the index tiny — the right starting point for laptop-class RAG prototypes, with e5-base-v2 the next step when quality tops out.