EmbeddingGemma 2 paper describes an open model for text, images, audio and video

Submitted Oct. 27, the EmbeddingGemma 2 paper describes an open model that represents text, images, audio, video and combinations of them in one embedding space. Co-author Omar Sanseviero shared the paper Nov. 3. The model has selectively loadable vision and audio encoders, with search among its intended uses. Its code-benchmark comparison with EmbeddingGemma 1 rests on developer-reported scores.