EmbeddingGemma
EmbeddingGemma is a Gemma 3-based open text embedding model under 500M parameters. It is the top open multilingual embedder of its size on MTEB and runs in under 200MB RAM quantized.
Trained on 100+ languages for on-device retrieval and RAG with customizable output dimensions. Part of the Gemma push toward small specialized models.
- Date
- Thursday, 4 September 2025
- Lab
- Google DeepMind
- Kind
- open-weights
- Access
- open weights (restricted license)
Figures
| Measure | Value | Measured by |
|---|---|---|
| MTEB ranking | top open multilingual text embedder under 500M company claim | company |
Sources
This record was checked against its sources on 6 October 2026. How we check