EmbeddingGemma 2
EmbeddingGemma 2 is Google's open embedding model, published October 6, 2026 under the Apache 2.0 license.
Provider
Model family
Gemma
Multimodal embedding model
Cost tier
Embedding
Status
Current
Release Oct 6, 2026
Why teams choose it
Text-only use can drop the vision and audio encoders and run the 270M text encoder.
Quantized on-device RAM cited on the post is about 191MB for text-only and about 567MB
for the full model on Pixel 11 Pro.
Tradeoffs to know
- Not a chat or image-generation model, and not a Gemini API model ID.
- Model Garden availability was still listed as coming soon on the launch post.
- The MTEB Code score on the launch post is vendor-reported.
Technical specs
- Inputs
- text, image, video, audio
- Outputs
- embedding
- Capabilities
- Text, image, video, and audio embeddings, 768-dimension vectors, truncatable to 512, 256, or 128, 8,000-token context, Apache 2.0 weights on Hugging Face and Kaggle
- License
- Apache 2.0
- Context tokens
- 8,000
- Parameters Text
- 270M
- Parameters Audio
- 300M
- Parameters Total
- 740M
- Parameters Vision
- 170M
- Embedding Dimensions
- 768, 512, 256, 128
Benchmarks
- Mteb Code (vendor-reported)
- 78.68
- Mteb Code Previous (vendor-reported)
- 68.76
Related models
Gemma 2 27B
Gemma 2 27B is Google’s open-weights Gemma family checkpoint balancing quality and deployability for research and pro…
Gemini 3.1 Pro Preview
Gemini 3.1 Pro Preview is Google's preview Gemini 3-series model for complex multimodal reasoning, software engineeri…
Gemini 2.0 Flash
Gemini 2.0 Flash is Google’s efficiency-oriented multimodal model generation aimed at fast agentic and interactive ex…
EmbeddingGemma 2 FAQ
What is EmbeddingGemma 2?
0 license. It is not a generative Gemini API model. The launch post says it has 740 million parameters: a 270 million text encoder, an optional 170 million vision encoder, and a 300 million audio encoder. Embeddings are 768 dimensions and can be truncated to 512, 256, or 128. Context is 8,000 tokens, four times EmbeddingGemma 1. 5 minutes of audio, 29 images, or 58 video frames, and that the text-only configuration can run at 270 million parameters. Weights are on Hugging Face and Kaggle. Google says a Model Garden listing is coming soon. 68.
When does EmbeddingGemma 2 fit best?
On-device and open-weight search, clustering, and retrieval
What should teams watch out for with EmbeddingGemma 2?
Not a chat or image-generation model, and not a Gemini API model ID.
Explore next
Models, tools, and comparisons that connect to this reference.