EmbeddingGemma 2: An Open, Lightweight Multimodal Embedding Model
Google, Tuesday, October 6th, 2026
Google released EmbeddingGemma 2, a 740M-parameter open model that embeds text, code, images, audio and video on-device.
Google launched EmbeddingGemma 2, extending its on-device embedding model beyond text to unify code, images, video and audio in one embedding space under an Apache 2.0 license.
Built on the Gemma 4 architecture, it has 740 million parameters with a 270M text-only core and optional vision and audio encoders, supports Matryoshka truncation from 768 to 128 dimensions and an 8K-token context, and runs in about 191MB of RAM for text on a Pixel 11 Pro.
Google reports a 9.92-point gain on MTEB Code over the original model, which had more than 20 million downloads.