EmbeddingGemma
by Google
· Gemma family · released Sep 4, 2025
Open weights
Active
EmbeddingGemma is Google's 308M-parameter open multilingual text embedding model, designed for on-device RAG and search and released in collaboration with the SentenceTransformers ecosystem.
This model has no first-party API. Open-weight release: self-host from published weights.
Specifications
| Model type | Embedding |
|---|---|
| Architecture | Dense |
| Parameters | 0.3 |
| Context window | 2.0K tokens |
| Max output | Not disclosed tokens |
| Knowledge cutoff | Not disclosed |
| License | Gemma Terms of Use |
| Modalities | Text input |
| Tags | Small LM |
| End of life | Not disclosed |
| Last verified | Not disclosed |