e5-large-v2

Large E5 embedding checkpoint for English semantic search, passage retrieval, and similarity tasks that follow the E5 query and passage format.

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

Teams that want a strong English retrieval baseline for search and RAG before moving to multilingual or larger embedding experiments.

Who should use it

Teams that want a strong English retrieval baseline for search and RAG before moving to multilingual or larger embedding experiments.
Builders who want local or self-hosted testing options.
RAG builders comparing retrieval quality, latency, and multilingual coverage.

Common workflows

English semantic search and RAG
embedding workflows
rag workflows
semantic search workflows
retrieval workflows

Deployment and hardware notes

~1.34 GB in fp32 (335M parameters), ~0.67 GB in fp16 — CPU suffices for prototypes; interactive query rates want a modest GPU.

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.
Teams that want a strong English retrieval baseline for search and RAG before moving to multilingual or larger embedding experiments.
Prefix every search string with 'query: ' and every document with 'passage: '; for symmetric similarity tasks the card says to use 'query: ' on both sides.

Limitations

English only and capped at 512 tokens per input; skipping the mandatory 'query: '/'passage: ' prefixes silently degrades ranking rather than erroring.
~1.34 GB in fp32 (335M parameters), ~0.67 GB in fp16 — CPU suffices for prototypes; interactive query rates want a modest GPU.
Context window and limits: 512 tokens.
Verify the exact model card, provider docs, license, and serving support before production use.

Local workflow notes

Prefix every search string with 'query: ' and every document with 'passage: '; for symmetric similarity tasks the card says to use 'query: ' on both sides.

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

Platforms: Windows, macOS, Linux

Embedding spec

Memory~1.34 GB in fp32 (335M parameters) · ~0.67 GB in fp16Output1024-dimension vectorsMax input512 tokens

The strongest English v2 E5: 1024-dimension mean-pooled vectors. Pick e5-base-v2 or e5-small-v2 when index size or CPU latency matters more than the last few points of ranking quality.