GoKawiilGoogle releases EmbeddingGemma 2, adding multimodal embeddings on-device
Google launched EmbeddingGemma 2, a follow-up to its original EmbeddingGemma model that reportedly surpassed 20 million downloads. The new 740-million-parameter model is built on the Gemma 4 architecture, released under an Apache 2.0 license, and extends beyond text to embed code, images, video, and audio in a single shared space for on-device inference.
GoKawiil's interpretation of the reporting above, not reported fact.
By unifying multiple data types into one embedding model that runs on consumer hardware, Google could make it easier for developers to build cross-modal search and retrieval tools without relying on cloud servers, which may appeal to privacy-conscious applications. The permissive licensing and compact size suggest Google is aiming to keep EmbeddingGemma's developer momentum going as multimodal AI features become more common in apps.
- EmbeddingGemma 2 adds support for code, images, video, and audio alongside text in one embedding space.
- The model has 740 million parameters and is built on the Gemma 4 architecture, licensed under Apache 2.0.
- It follows the original EmbeddingGemma, which Google says exceeded 20 million downloads.
Source: blog.google — Sahil Dua, 2026-10-06
Published there as: “EmbeddingGemma 2”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.