e5-small

e5-small is an E5-family embedding model for semantic search, multilingual retrieval, low-overhead indexing, or RAG pipelines.

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

Legacy lightweight embedding baseline

Who should use it

Legacy lightweight embedding baseline
Builders who want local or self-hosted testing options.
RAG builders comparing retrieval quality, latency, and multilingual coverage.

Common workflows

Legacy lightweight embedding baseline
embedding workflows
rag workflows
semantic search workflows
retrieval workflows

Deployment and hardware notes

~0.13 GB in fp32 — runs anywhere, including modest laptops with no GPU.

License and usage notes

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

Strengths

Open weights where released model option for E5 workflows.
Legacy lightweight embedding baseline
Requires the 'query: '/'passage: ' prefixes like all E5 models; forgetting them degrades results silently.

Limitations

Legacy v1 small: same 33.4M parameters and 384 dimensions as e5-small-v2, which simply ranks better. English only, 512 tokens. The gap to v2 widens on question-style queries, where the retraining effort concentrated.
~0.13 GB in fp32 — runs anywhere, including modest laptops with no GPU.
Context window and limits: 512 tokens.
Verify the exact model card, provider docs, license, and serving support before production use.

Local workflow notes

Requires the 'query: '/'passage: ' prefixes like all E5 models; forgetting them degrades results silently.

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

The original tiny E5, superseded by e5-small-v2 in the same footprint. Migrating requires re-embedding — v1 vectors and v2 vectors are not comparable. If CPU budget is the constraint that led you here, the v2 checkpoint costs exactly the same to serve.