EmbeddingGemma 2

EmbeddingGemma 2 is Google’s October 6, 2026 open multimodal embedding model, released as google/embeddinggemma-2 under Apache 2.0.

EmbeddingsLocal RetrievalApache 2.0Google

Local Sentence Transformers-compatible deployment; device/runtime requirements vary by encoder

Apache 2.0 open weights; local compute costs apply, no hosted token rate established here.

Commercial

🚀Function Overview

EmbeddingGemma 2 is Google’s October 6, 2026 open multimodal embedding model, released as google/embeddinggemma-2 under Apache 2.0.

Key Features

Official downloadable ID google/embeddinggemma-2; Apache 2.0
740M full model, modular text/vision/audio encoders
Text/code, image, video and audio input; numerical embeddings output
Shared 8,192-token input context
768 native dimensions; 512/256/128 truncation supported
Re-normalize truncated vectors; keep query/corpus dimensions equal

Use Cases

•Local cross-modal semantic search.
•Retrieve images, video moments or audio using a query.
•Embed a local corpus for RAG.
•Zero-shot classification and clustering with reviewed labels.

⚙️Input Parameters

sentences

array

Text or batch of texts passed to SentenceTransformer("google/embeddinggemma-2").encode; multimodal inputs use the dedicated encoder contract.

prompt_name

string

Use the documented query or document prompt consistently with the retrieval task.

truncate_dim

integer

768, 512, 256 or 128; keep query/document dimensions equal.

normalize_embeddings

boolean

Normalize embeddings after dimension truncation.

💡Usage Examples

Example 1

Input Parameters

{
  "sentences": [
    "A query about a local image collection"
  ],
  "prompt_name": "query",
  "truncate_dim": 256,
  "normalize_embeddings": true
}

Output Results

Unexecuted encode keyword-argument example after initializing SentenceTransformer("google/embeddinggemma-2"). A one-item batch produces shape (1, 256); numeric values require actual inference.

Quick Actions

Technical Specifications

Hardware Type: Local Sentence Transformers-compatible deployment; device/runtime requirements vary by encoder
Commercial Use: Supported
Pricing: Apache 2.0 open weights; local compute costs apply, no hosted token rate established here.