Gemma 4

Latest news

1. 7 Oct
   Google DeepMind releases EmbeddingGemma 2, a multimodal on-device embedding model

   Google DeepMind announced EmbeddingGemma 2 on Oct 6, 2026, a 740M-parameter multimodal embedding model that maps text, images, audio, video and code into a shared embedding space. It is built on the Gemma 4 architecture, intended for on-device inference, and released under the Apache 2.0 license.

   Models · 2 sources

   

2. 6 Oct
   Transformers v5.19 adds EmbeddingGemma2; Polars 2.0 ships SQL and out-of-core

   Transformers v5.19.0 adds a new multimodal embedding model, EmbeddingGemma2, with configurable token budgets and truncated embedding sizes. Separately, Polars 2.0 launches with out-of-core (spill-to-disk) support, first-class SQL, performance improvements, a new Map dtype, and stricter dtype behavior.

   Tools · 2 sources

   

Articles

No articles about Gemma 4 yet.