Google releases EmbeddingGemma 2, an open multimodal embedding model

Archive entry: this covers an event from October 6, 2026, and we published it on November 8, 2026. The details come from sources dated at the time and may have changed since.

On October 6, 2026, Google released EmbeddingGemma 2, an open multimodal embedding model, under the Apache 2.0 license. Google described the release in its announcement on The Keyword blog.

What the model offered

• 740M parameters, built on the Gemma 4 architecture, embedding text, code, images, video and audio in a single embedding space.
• Text-only use started at 270M parameters, with separate 170M vision and 300M audio encoders.
• Matryoshka (MRL) truncation reduced embeddings from 768 dimensions to 512, 256 or 128, cutting storage by up to 6x.
• An 8K token context window, 4x the first version, which fit 5.5 minutes of audio, 29 images or 58 video frames.
• On a Pixel 11 Pro, the text-only model used about 191 MB and the full model about 567 MB.
• The MTEB Code score rose from 68.76 to 78.68.

What it meant for people who build with AI

Developers got an open, on-device multimodal embedding model for retrieval and RAG. Because the weights were published under Apache 2.0, teams could download the model and run it themselves, and the Matryoshka sizes let them shrink embeddings from 768 to as few as 128 dimensions to save storage.

Availability and background

The weights were published on Hugging Face and Kaggle. Google noted that the first EmbeddingGemma had more than 20M downloads.

Sources

• EmbeddingGemma 2: an open, lightweight multimodal embedding model (Google (The Keyword), October 6, 2026, official)

Written with AI assistance, strictly from the sources listed above. Before it went live, automated checks verified its dates, sources and every language edition.