What changed
Google’s Gemma release log records EmbeddingGemma 2 at 740 million parameters. The same log separates it from earlier Gemma 4 releases, helping readers distinguish model families and purposes.
The everyday takeaway
Embeddings turn material into numerical representations useful for tasks such as finding related text. A model that helps retrieve information can be important even when it does not write the final answer. For example, a search feature can find a relevant help article first and then pass that article to a separate assistant.
What we have verified
Google’s release log lists EmbeddingGemma 2 on October 6. We have not run it. This is separate from the Gemma 4 language-model family.