Google Releases Compact EmbeddingGemma 2 Model

Google released EmbeddingGemma 2, an open model with 740 million parameters that outperforms rival models twice its size on benchmarks. It runs locally via WebGPU in 20 to 70 milliseconds, needs 191 MB of RAM, and cuts storage by up to six times. The weights are available on Hugging Face and Kaggle.