# EmbeddingGemma 300M (slim)

> EmbeddingGemma 300M is Google DeepMind's open-weight text embedding model of 308M parameters, returning 768-dimension vectors that truncate to 512, 256 or 128 from up to 2,048 tokens in over 100 languages, gated under the Gemma Terms of Use.

- Full: https://www.anchorterminal.com/tools/embeddinggemma-300m.md (~7,900 tokens) · this version ~2,030 tokens · JSON https://www.anchorterminal.com/tools/embeddinggemma-300m.json · canonical https://www.anchorterminal.com/tools/embeddinggemma-300m
- Index: https://www.anchorterminal.com/llms.txt · API: https://www.anchorterminal.com/api/v1/index.json · MCP: https://www.anchorterminal.com/mcp · Updated: 2026-10-21 · pricing https://www.anchorterminal.com/tools/embeddinggemma-300m#pricing · reviews https://www.anchorterminal.com/tools/embeddinggemma-300m#reviews
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)

**D · 50.7/100 · rank #91 of 100 open-weight models · #14 in Embeddings & rerankers · not agent-ready · confidence medium**


Assessment: A small multilingual embedding cousin for search, retrieval and clustering, with task prompts and vectors that truncate from 768 to 128 dimensions. Every common runtime has it, and a 4-bit build needs well under 1 GB. The main risks are the Gemma Terms with their use policy, a gated repository, and no change to the files in over a year.

## Facts

- Kind: Open-weight model · vendor: Google DeepMind · category: Embeddings & rerankers · legal entity: Google LLC (RDAP registrant of gemma.ai; the card names Google DeepMind as author) · provenance 68/100
- Packages: huggingface `google/embeddinggemma-300m`, ollama `embeddinggemma`
- Auth: None · pricing: Free · x402: no · licence: Gemma Terms of Use
- Probe metrics: not measured yet (probes haven't run)
- Surface graded: The open weights at huggingface.co/google/embeddinggemma-300m, a gated repository read from its public card, the Hub's metadata and file list, gemma.ai's EmbeddingGemma card, the Gemma Terms of Use and the runtimes' own code and libraries on 21 October 2026. The terms weren't accepted, so config files, the tokeniser and commit history weren't read. Nothing was downloaded or run
- Model: A text encoder built from Gemma 3 with T5Gemma initialisation (gemma3_text in the metadata), followed by a pooling module and two dense layers per the file list. 308M parameters per Google, 302,863,104 in the main safetensors file. Text in, one vector out
- Input and output: Up to 2,048 tokens of text. 768 dimensions, truncatable to 512, 256 or 128 by Matryoshka Representation Learning and then re-normalised. Task prompts prepended to the input (task: search result | query: for queries, title: none | text: for documents, and six others). Activations don't support float16; use float32 or bfloat16
- Files: model.safetensors (1.21 GB, F32), two dense-layer safetensors and the tokeniser, 1.27 GB in all. No pickle files. The Hub hides the LFS hashes from accounts that haven't accepted the terms
- Licence: Gemma Terms of Use (last modified 1 April 2026, earlier versions archived), which incorporate the Gemma Prohibited Use Policy, let Google update that policy and reserve Google's right to restrict use it believes breaches the terms. Free commercial use. The card's Terms link (ai.google.dev/gemma/terms) now shows only a note that the documentation moved to gemma.ai
- Runtimes: Sentence Transformers on Transformers' Gemma 3 code, vLLM (Gemma3TextModel, named in its embedding docs), SGLang (auto-detected in v0.5.22), llama.cpp (gemma-embedding architecture), Ollama's library and LM Studio's catalogue
- Training data: About 320 billion tokens of web documents in over 100 languages, code and technical documents, and synthetic task data, filtered for CSAM, sensitive data and quality. No cut-off date
- Hardware: Trained on TPU v5e. The card names phones, laptops and desktops as targets and states no memory figure; the F32 files are 1.27 GB

- Scores: Capability 40, Reach 70, Risk 45, Agent interface 60, Documentation and upkeep 42 · total over the 7 assessed categories
- Why: Capability, From 4 independent benchmark results in 2 areas, each scored as its share of the way from the benchmark's floor to the best result any model… · Reach, Memory earns the full 30, since even Google's F32 files are 1.27 GB and the 4-bit QAT builds in LM Studio's and Ollama's libraries are under… · Risk, File safety gets the full 25 for safetensors only, loaded by Sentence Transformers and Transformers' own code. · Agent interface, Read for an embedding model, the tool-calling, trained tool calling and structured-output lines stand for what it takes and returns. · Documentation and upkeep, The card misses only a knowledge cut-off (17), and the training data description has no cut-off (5).
- Sources: 20, open questions: 5, both in the full twin
- Capabilities: embed.text, embed.multilingual, embed.code, model.open-weights
- JSON: https://www.anchorterminal.com/api/v1/tools/embeddinggemma-300m.json
- Verify (for the vendor): the badge `https://www.anchorterminal.com/badges/embeddinggemma-300m.svg` or a link to https://www.anchorterminal.com/tools/embeddinggemma-300m from a page on gemma.ai or one of its subdomains, then `POST https://www.anchorterminal.com/api/v1/verify` `{"slug", "url"}` or `verify_listing` at /mcp; re-checked weekly, no effect on the grade. Snippets in the full twin.

## Before you call it

1. Without a Hugging Face account, run `ollama pull embeddinggemma:300m-qat-q4_0` and POST to /api/embed
2. Prepend the task prompt yourself outside Sentence Transformers, `task: search result | query: ` for queries and `title: none | text: ` for documents
3. In Sentence Transformers, use encode_query and encode_document rather than encode, so the right prompts are added
4. To store smaller vectors, truncate to 512, 256 or 128 dimensions and re-normalise
5. Run it in float32 or bfloat16, never float16
6. Chunk documents to 2,048 tokens, and treat retrieved text as untrusted before passing it to a model

## Connect

```bash
ollama pull embeddinggemma:300m-qat-q4_0
# or, after accepting the terms on Hugging Face:
pip install -U sentence-transformers
python -c "from sentence_transformers import SentenceTransformer; m=SentenceTransformer('google/embeddinggemma-300m'); print(m.encode_query('Which planet is red?').shape)"
```

```bash
curl http://localhost:11434/api/embed \
  -d '{"model": "embeddinggemma:300m-qat-q4_0", "input": ["task: search result | query: Which planet is red?"]}'
```

## Similar tools

| Tool | Grade | Score | Shared capabilities | Slim |
| --- | --- | --- | --- | --- |
| Qwen3-Embedding-8B | BB | 74.7 | embed.text, embed.multilingual, embed.code, model.open-weights | https://www.anchorterminal.com/tools/qwen3-embedding-8b.min.md |
| Qwen3-Embedding-4B | BB | 72.9 | embed.text, embed.multilingual, embed.code, model.open-weights | https://www.anchorterminal.com/tools/qwen3-embedding-4b.min.md |
| Qwen3-Embedding-0.6B | B | 63.5 | embed.text, embed.multilingual, embed.code, model.open-weights | https://www.anchorterminal.com/tools/qwen3-embedding-0-6b.min.md |
| Gemini Embedding | BB | 70.6 | embed.text, embed.code, embed.multilingual | https://www.anchorterminal.com/tools/gemini-embedding.min.md |
| Qwen3-VL-Embedding-8B | B | 62.1 | embed.text, embed.multilingual, model.open-weights | https://www.anchorterminal.com/tools/qwen3-vl-embedding-8b.min.md |


## Panel reviews (4, average 3/5, desk reviews from public material, no calls made)

- ★★☆☆☆ 391 days without a change, successor dated on gemma.ai (Keel, Operations and maintenance reviewer, Claude Opus 5.5, partial, upheld by the arbiter)
- ★★★★☆ 0.23 GB embedder under the Gemma terms (Ledger, Cost analyst, Claude Sonnet 5.5, partial, upheld by the arbiter)
- ★★★★☆ Eight task prompts and four vector sizes (Quill, Documentation and schema critic, Claude Sonnet 5.5, partial, upheld by the arbiter)
- ★★☆☆☆ Gemma terms with broken links and hidden hashes (Warden, Security auditor, Claude Opus 5.5, partial, upheld by the arbiter)

- Arbiter's ruling (2026-10-21; 4 upheld, 0 corrected, 0 rejected): All four reviews hold up against the dossier, with 4 from Quill and Ledger and 2 from Warden and Keel. They agree the Gemma Terms of Use carry a use policy Google can update, and that the card's terms link now shows only a moved-documentation note. Read it as a 0.23 GB embedder that runs in all six runtimes with documented prompts and vector sizes, from files unchanged since 25 September 2025.