EmbeddingGemma 2 is Google DeepMind's open embedding model that puts text, code, images, video, and audio into one shared space. That means you can search across all of them with a single query, without juggling separate systems. It has 740M parameters, so it is light enough to run on-device. On a Pixel 11 Pro, the full multimodal model needs about 567MB of RAM, while the text-only version needs just 191MB. It is licensed under Apache 2.0, and the weights are available on Hugging Face and Kaggle. The model also shows a strong gain on the MTEB Code benchmark, with its score rising from 68.76 to 78.68.
Launch Date:
October 18, 2026