Google launches EmbeddingGemma 2, a multimodal AI embedding model

Google has released EmbeddingGemma 2, an open multimodal embedding model small enough to run on a smartphone. The update extends the original text-only EmbeddingGemma, introduced in September 2025, to images, audio, and video, all sharing a single embedding space. SiliconANGLE reported the launch today.

The move follows recent updates from Google, including the limited rollout of Gemini 4 Argon to cybersecurity teams earlier this month. EmbeddingGemma 2 appears to continue the trend of releasing models with specific use cases in mind.

The model’s design suggests potential for broader adoption, though its real-world impact will depend on developer uptake. Multimodal embeddings have traditionally been associated with cloud-based systems, and on-device implementations could present new opportunities—or challenges—for applications that rely on such capabilities.

The release comes amid ongoing developments in AI infrastructure, with companies exploring different approaches to model deployment. How EmbeddingGemma 2 fits into this landscape remains to be seen, particularly as Google maintains selective access to some of its other models.

Sources: siliconangle.com