Skip to content- Sources: primary
- Summary: Google published EmbeddingGemma 2 embedding weights under Apache 2.0, totaling 740M parameters, with a 270M text-only configuration plus optional 170M vision and 300M audio encoders. Google puts active RAM with quantization on a Pixel 11 Pro at about 191MB for text-only weights and about 567MB for the full multimodal model, and reports MTEB Code rising from 68.76 to 78.68, which is a vendor-run figure. Matryoshka truncation from 768 down to 128 dimensions cuts vector storage up to sixfold.
- Why it matters: On-device retrieval over text, images, audio, video and code fits a memory budget a phone or laptop holds, with an 8K token context window, four times EmbeddingGemma 1, under a license that permits commercial use.