Google's EmbeddingGemma 2 Claims Superior Performance Over Larger Models

Google's EmbeddingGemma 2, released on October 6, 2026, boasts 740 million parameters and achieves a score of 78.68 on the Massive Text Embedding Benchmark, surpassing its predecessor, Gemma 4, which scored 68.76. The new model requires only 191 MB of RAM and can reduce local vector database storage by up to six times. Query response times are reported to be between 20 to 70 milliseconds via WebGPU in the browser. Google asserts that EmbeddingGemma 2 outperforms competing models that are up to twice its size in multimodal embedding benchmarks. This development may influence practitioners to consider EmbeddingGemma 2 for applications requiring efficient storage and rapid query responses, especially in environments constrained by resource availability.