The development
What happened
Google announced EmbeddingGemma 2 on October 6, a lightweight embedding model for retrieving information across several media types on consumer hardware.
Key facts
The model has 740 million parameters, uses Gemma 4 architecture and is released under Apache 2.0.
Text, code, images, audio and video map into a shared embedding space.
Google reports an MTEB Code score increase from 68.76 to 78.68 over EmbeddingGemma.
It shares Gemma 4's text tokenizer and audio encoder, which Google says lowers the combined memory footprint.
The context
Why it matters
A shared local retrieval model can connect files and media without sending each query to a cloud API. It can support multimodal retrieval-augmented generation, but remains an embedding model. Benchmark results are publisher-reported; application design still determines privacy and practical performance.