Google releases Apache 2.0 EmbeddingGemma 2 weights and deployment paths.
Google releases EmbeddingGemma 2 for on-device multimodal retrieval
2026-10-07Verified announcementTRIAL
Google released the Apache 2.0 EmbeddingGemma 2 weights, a 740M-parameter model that maps text, code, images, video and audio into one 768-dimensional space. Its modality encoders can be loaded selectively, and Matryoshka training supports 128- to 768-dimensional outputs. Google reports 191 MB active RAM for text-only quantized weights and 567 MB for the full multimodal model on a Pixel 11 Pro.
What this means for you
For local search and RAG teams, one small open model can replace separate text, image and audio embedding stacks. Vendor benchmarks and device-memory claims still need validation on the target corpus and hardware.
Evidence & related material
Technical details & scores
| Parameters | 740M total; 270M text, 170M vision and 300M audio components |
|---|---|
| Modalities | text, code, image, video and audio |
| Context | 8K tokens; up to 5.5 minutes audio, 29 images or 58 video frames |
| Embedding Dimensions | 128, 256, 512 or 768 via Matryoshka truncation |
| License | Apache 2.0 |
| Reported Code Gain | MTEB Code 68.76 to 78.68 |
Technical impact 4
Engineering value 5
Adoption signal 3
Maturity 3
Verification cost 2
Risk 2