All models
Google DeepMind
EmbeddingGemma 2
Google DeepMind's open-weight embedding model for local search across text, code, images, audio, and video.
Plain-English overview
What EmbeddingGemma 2 actually is
EmbeddingGemma turns content into searchable vectors rather than generating an answer. Different media types can share one index, so a text query can retrieve a relevant picture, recording, or video segment.
Load only the encoders your application needs. Smaller vectors can reduce storage, but retrieval quality, preprocessing, and the actual device workload still need evaluation.
Good fit for
- Local and offline multimodal search
- RAG over private media and documents
- Codebase retrieval on consumer hardware
Category comparison
The facts that matter for embeddings models
These are provider-published specifications, not Cody benchmark scores. Follow the linked sources for current limits and endpoint-specific exceptions.
- Embedding price
- Open weights; self-hostedCurrent provider price per million input tokens or the closest published billing unit.
- Input capacity
- 8,192 shared tokens across text and mediaMaximum content accepted in one embedding input, using the provider's documented token basis.
- Vector dimensions
- 768 default; 512, 256, or 128 with re-normalizationSupported output sizes; smaller vectors reduce storage while larger vectors may preserve more information.
- Accepted inputs
- Text, code, images, audio, and sampled video framesText, code, image, audio, video, PDF, or document-aware input supported by the endpoint.
- Retrieval controls
- Task prompts, selective encoders, vector truncation, and normalizationQuery/document modes, task types, truncation, chunking, or other controls that shape vectors for retrieval.
- Where it runs
- Local or self-hosted; Apache 2.0 weightsDirect API, cloud marketplace, private deployment, or self-hosted route documented by the provider.
Pricing & comparisons
Estimate your cost
Set your usage. Your estimate updates as you type.
Assumes 600 tokens per page, processed separately. Actual token counts vary. This covers embedding only, not storage, search, or generated answers.
EmbeddingGemma 2
Google DeepMind
Estimated total (USD)
For the usage above · USD · API pricing, not a subscription
How this estimate works
Estimates exclude taxes, tools, cache storage/writes, free allowances and custom discounts. Image estimates cover output only, not prompt or reference-image charges. Quality modes differ by model. Unlisted settings are not treated as free.
API and provider access
Where to get EmbeddingGemma 2
Hugging FaceDownload official open weightsGoogle's Apache 2.0 checkpoint and model card.Google AI for DevelopersEmbeddingGemma documentationModel limits, task prompts, runtime guidance, and deployment options.Google AI EdgeTry the on-device search demosOfficial local media-search and video-retrieval examples.Cody editorialRead the EmbeddingGemma 2 guideA practical explanation of local retrieval, version changes, and model tradeoffs.
Availability
Regions and access stage
Open weights for local or self-hosted inference, with optional vision and audio encoders.
Download access follows the hosting platform's policies. Self-hosted deployment is not a provider-operated regional API.
Check live availabilityData and training
The route matters.
Inference can run locally without sending source content to an embedding API. Data retention and network access depend on the application and services you connect.
This is a concise reading of the cited provider material, not legal advice. A third-party gateway can have different storage, routing, training, and residency terms from the model maker's direct API.
Read the provider policyFrequently asked questions
EmbeddingGemma 2 FAQ
What is EmbeddingGemma 2?
Google DeepMind's open-weight embedding model for local search across text, code, images, audio, and video. EmbeddingGemma turns content into searchable vectors rather than generating an answer. Different media types can share one index, so a text query can retrieve a relevant picture, recording, or video segment.
When was EmbeddingGemma 2 released?
EmbeddingGemma 2 was released on October 6, 2026 according to the cited provider materials.
Where can I access EmbeddingGemma 2?
Open weights for local or self-hosted inference, with optional vision and audio encoders. The access routes listed in this guide are Hugging Face, Google AI for Developers, Google AI Edge, and Cody editorial.
How much does EmbeddingGemma 2 cost?
Open weights; self-hosted. Apache 2.0 weights. Compute, memory, and hosting costs depend on your deployment; the release documentation does not list a fixed hosted API token rate.
Where is EmbeddingGemma 2 available?
Open weights for local or self-hosted inference, with optional vision and audio encoders. Download access follows the hosting platform's policies. Self-hosted deployment is not a provider-operated regional API.
Is my EmbeddingGemma 2 API data used for training?
Inference can run locally without sending source content to an embedding API. Data retention and network access depend on the application and services you connect. The policy belongs to the provider route and account terms, so verify it again before production use.
Price comparison
USD per million tokens for the shown API routes at standard context length. Cache and long-context rates may differ. Models without matching reviewed prices are omitted; lower cost does not mean better quality.
- Mistral Embed$0.10
Mistral AI
- Voyage 4 Large$0.12
Voyage AI
- Voyage Context 4$0.12
Voyage AI
- Voyage Code 4$0.12
Voyage AI
- text-embedding-3-large$0.13
OpenAI
- Gemini Embedding 2$0.20
Google
Related comparisons
Head-to-head comparisons
EmbeddingGemma 2 vs Gemini Embedding 2
Google's multimodal embedding model for placing text, images, video, audio, and PDFs in one searchable vector space.
Open full comparisonEmbeddingGemma 2 vs text-embedding-3-large
OpenAI's highest-capability text embedding model for semantic search, recommendations, clustering, and retrieval pipelines.
Open full comparisonEmbeddingGemma 2 vs Cohere Embed 4
Cohere's enterprise embedding model for multilingual text, images, and visually rich documents with a 128K context window.
Open full comparison
