Embedding

Gemma Terms of UseOpen weights where releasedUpdated August 2026

bge-embedding-gemma2

bge-embedding-gemma2 is a BGE-family model for embeddings, reranking, retrieval, semantic search, or RAG workflows.

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

Gemma-based embedding experiments

Who should use it

  • Gemma-based embedding experiments
  • Builders who want local or self-hosted testing options.
  • RAG builders comparing retrieval quality, latency, and multilingual coverage.

Common workflows

  • Gemma-based embedding experiments
  • embedding workflows
  • rag workflows
  • retrieval workflows
  • semantic search workflows

Deployment and hardware notes

Needs a serious GPU: ~18.5 GB in fp16 means a 24 GB card for comfortable serving; quantized serving cuts that, but embedding quality under quantization is your benchmark to run.

License and usage notes

Gemma Terms of Use. 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 BGE workflows.
  • Gemma-based embedding experiments
  • Budget throughput carefully — bulk-indexing a large corpus through a 9.2B model is orders of magnitude more compute than this page's sub-1B siblings.

Limitations

  • LLM-scale serving costs for an embedder: every document embedding is a 9.2B-parameter forward pass, and the 3584-dimension vectors make indexes several times larger than 1024-dimension alternatives. The Gemma license adds use-restriction terms MIT-licensed rivals do not carry.
  • Needs a serious GPU: ~18.5 GB in fp16 means a 24 GB card for comfortable serving; quantized serving cuts that, but embedding quality under quantization is your benchmark to run.
  • Context window and limits: 8,192 tokens.
  • Verify the exact model card, provider docs, license, and serving support before production use.

Local workflow notes

Budget throughput carefully — bulk-indexing a large corpus through a 9.2B model is orders of magnitude more compute than this page's sub-1B siblings.

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

Platforms: Windows, macOS, Linux

Embedding spec

Memory~18.48 GB in fp16 (9.2B parameters) — an LLM-scale embedding model · ~5.6 GB at Q4_K_MOutput3584-dimension vectorsMax input8,192 tokens

Gemma-2-based multilingual embedder emitting 3584-dimension vectors with an 8,192-token window. Reach for it only when measured quality gains over bge-m3 justify roughly 16x the serving compute.

Sources to verify

Exact model card

Related resources

Continue with model source notes, local tools, and implementation guides related to this model.

BGE family hubEmbedding models shortlistCompare BGE vs E5RAG for documents stackAI tools directory
Hardware~18.48 GB in fp16 (9.2B parameters) — an LLM-scale embedding modelRuntimeOllama or LM Studio where supported, llama.cpp, Transformers, vLLMContext8,192 tokensLast updated2026
Exact model card →

Model ecosystem connections

Use these next-step links to move from this profile into related tools, comparisons, guides, stacks, and curated shortlists.

Recommended runtimes and tools

Ollama →LM Studio →llama.cpp →Transformers →vLLM →

Setup and deployment

Check if this model fits your GPU →What can 24GB VRAM run? →What is quantization? GGUF, Q4, Q8 explained →What is VRAM and how much do you need? →

Related model pages

BGE family hub →bge-large-en-v1.5 →bge-large-zh-v1.5 →bge-m3 →bge-base-en-v1.5 →bge-reranker-base →bge-reranker-large →

Guides, stacks, and comparisons

Embedding models shortlist →Compare BGE vs E5 →RAG for documents stack →All model comparisons →

Related models

bge-large-en-v1.5English retrieval stacks that want a stronger embedding baseline for search and RAG before moving to multilingual models.bge-large-zh-v1.5Chinese embedding and retrieval workflowsbge-m3RAG teams that want one embedding model for multilingual search, long documents, and retrieval experiments beyond plain dense vectors.bge-base-en-v1.5Teams building English search or RAG systems on modest GPUs without dropping to the smallest embedding tier.bge-reranker-baseBuilders who want a practical reranking step in RAG or search without adding a heavyweight cross-encoder to every request.bge-reranker-largeTeams that already retrieve a candidate set and want a stronger final ranking step before passing context to an LLM.

Ready to run this model locally?

Find a compatible interface in our Local AI Tools directory →