Google DeepMind Releases EmbeddingGemma 2 (740M)
Google DeepMind launches EmbeddingGemma 2, a 740M open multimodal embedding model built on Gemma 4 supporting 5 inputs under Apache 2.0.
Stock photo for illustration only, not from the actual event
- Google DeepMind officially released EmbeddingGemma 2 on October 6, 2026
- A 740 million parameter (740M) open multimodal model built on Gemma 4 architecture
- Maps 5 distinct input types into a unified 768-dimensional (768d) vector space
- Available for immediate use under the open Apache 2.0 license framework
The artificial intelligence community receives a major boost as Google DeepMind announces the release of EmbeddingGemma 2, a new open multimodal embedding model boasting 740 million parameters. Built directly upon the robust foundation of the Gemma 4 architecture, this model becomes available to developers and researchers starting October 6, 2026.
A core breakthrough of EmbeddingGemma 2 lies in its capability to process five separate input types, mapping all of them into a single, cohesive 768-dimensional vector space, commonly referred to as the 768d space. This advanced architectural design enables the system to seamlessly bridge and understand cross-modal relationships among diverse data formats with remarkable efficiency.
Rather than restricting access to closed environments, Google DeepMind has shipped EmbeddingGemma 2 under the permissive Apache 2.0 open-source license. This empowers developers worldwide to integrate, modify, and deploy the model for both commercial and research applications with broad operational freedom.
Releasing EmbeddingGemma 2 under the Apache 2.0 license underscores Google DeepMind's commitment to accelerating open-source AI ecosystems. By consolidating five input types into a uniform 768d vector space, the model drastically simplifies the development pipelines for multimodal Retrieval-Augmented Generation (RAG) systems, allowing smaller engineering teams to leverage enterprise-grade capabilities.
Stock photo for illustration only, not from the actual event
The arrival of EmbeddingGemma 2 on top of the Gemma 4 framework further demonstrates Google's ongoing dedication to expanding its Gemma model family, prioritizing both high performance and deployment flexibility across diverse computing environments ranging from edge devices to cloud infrastructures.
Source: MarkTechPost
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment