The EmbeddingGemma 2 team on Tuesday said it released EmbeddingGemma 2, an open embedding model that can handle more than text.
The team said the release adds built-in ability to understand several kinds of media to an open embedding model that previously handled only text. It places written words, pictures, sound and video, or mixes of those inputs, in one shared embedding space. The team also said it compressed the model backbone into a more compact profile.
The post did not say the model’s size, price or where developers can obtain it.
Google published the announcement on its Keyword site and pointed readers to an official technical report. The team said the report gives in-depth information on the training strategy, evaluation methodology and experimental results.
The team said the model’s recipe includes innovations that build on technology from Gemma 4 and Gemini Embedding 2. Those changes, the team said, enable EmbeddingGemma 2 to learn “an expressive and unified vector space.”
The post also invites readers to watch a five-minute explainer on how EmbeddingGemma 2 works and how it was developed. It did not say what the explainer covers beyond that description.