Google Deepmind Unveils 740M-parameter Multimodal Model

WHO:

MarkTechPost - Google DeepMind is the organization releasing the model.

Techmeme - Google is the parent company of Google DeepMind.

WHAT:

MarkTechPost - Google DeepMind has released EmbeddingGemma 2, a 740M-parameter open multimodal embedding model.

Techmeme - EmbeddingGemma 2 is designed to map code, images, video, and audio into a shared embedding space.

Techmeme - The model is described as the most capable for on-device multimodal embeddings.

WHERE:

MarkTechPost - The model is available under the Apache 2.0 license.

WHEN:

MarkTechPost - The model ships today.

HOW:

MarkTechPost - EmbeddingGemma 2 maps five input types into one 768-dimensional space.

(UPDATED October 6, 2026 19:09 UTC)

Google DeepMind has released a new model called EmbeddingGemma 2, which features 740 million parameters and is designed to map code, images, video, and audio into a shared embedding space. This model is touted as the most capable for on-device multimodal embeddings and is available under the Apache 2.0 license. EmbeddingGemma 2 effectively maps five input types into a single 768-dimensional space and is shipping today.