Google DeepMind launches EmbeddingGemma 2, a 740M-parameter multimodal embedding model

Google DeepMind has released EmbeddingGemma 2, a 740M-parameter model that maps code, images, video, and audio into a shared embedding space. It ships under an Apache 2.0 license and is billed as the most capable model for on-device multimodal embeddings.

• 740M parameters, natively maps text, images, video, and audio
• Shared embedding space for code, images, video, and audio
• Released under the Apache 2.0 license
• Claimed as the most capable on-device multimodal embedding model