Google's Gemini Embedding 2 arrives with native multimodal support to cut costs and speed up your enterprise data stack
Google's Gemini Embedding 2 offers native multimodal support, reducing costs and latency for enterprise AI tasks by unifying data types into a single embedding space.Summary
Google has released a public preview of Gemini Embedding 2, a new embeddings model that natively integrates text, images, video, audio, and documents into a single numerical space. This advancement significantly improves efficiency for enterprise AI applications, reducing latency by up to 70% and lowering costs for companies utilizing AI powered by their own data. Unlike previous models that often require converting media into text, Gemini Embedding 2 understands different data types directly, minimizing translation errors and capturing nuanced information.
The model utilizes a 3,072-dimensional space and a technique called Matryoshka Representation Learning, allowing enterprises to balance precision with storage costs by adjusting the vector dimensions. Benchmarks demonstrate Gemini Embedding 2's superior performance in multimodal retrieval, speech and audio depth, and contextual scaling compared to existing industry leaders.
Early adopters like Sparkonomy and Everlaw have reported substantial gains in efficiency and accuracy. Google offers Gemini Embedding 2 through the Gemini API for prototyping and Vertex AI for large-scale deployments, with tiered pricing and integration with popular AI infrastructure tools. The move towards natively multimodal embeddings represents a significant step towards a unified knowledge base for enterprises, enabling more advanced Retrieval-Augmented Generation (RAG) and deeper insights from diverse datasets.
(Source：Venturebeat)