EmbeddingGemma 2
·
A compact model that maps different media into a shared embedding space for local search and retrieval. Text, vision and audio components can be selected to fit the application.
- Reviewed tasks
- Search and retrieval
- Reviewed modalities
- Text, Images, Audio, Video
- Deployment paths
- Downloadable weights
- Focus / inputs
- Text, images, audio and video → embeddings
- Availability / restrictions
- Downloadable weights on Hugging Face and Kaggle, with local deployment tools linked from the announcement.
- Licence category
- Apache-2.0
- Licence / terms
- Apache 2.0, as stated in the official release.
Variants, limits and list prices
Values are copied from the linked first-party pages as stated, with the date each group was checked. Unknown means the pages checked do not state the value. Prices are list prices, not a quote or a ranking.
| Field | EmbeddingGemma 2 |
|---|---|
| API identifier | Unknown |
| Released | Unknown |
| Status and retirement | Unknown |
| Lifecycle checked | Checked |
| Context window | 8,192 tokens shared across modalities |
| Max input | Unknown |
| Max output | Not applicable: 768-dimension embeddings, with Matryoshka sizes 128, 256 and 512 |
| Rate limits | Unknown |
| Limit notes | Unknown |
| Limits checked | Checked |
| Base input price | No list price |
| Base output price | No list price |
| Cache-read price | No list price |
| Price conditions | Free downloadable weights; no usage list price applies. |
| Prices checked | Checked |
| Knowledge cutoff | Unknown |
| Cutoff checked | Checked |
Parameters and weights
- Parameters
- 740M total: 270M text (130M backbone and 140M embedder), 170M vision and 300M audio
- Formats
- Safetensors (BF16)
- Checked