What happened
On October 6, 2026, Google DeepMind rolled out EmbeddingGemma 2, expanding past text so one shared embedding space covers code, pictures, clips, and sound. Weights went live on Hugging Face and Kaggle at launch under the Apache 2.0 license, which is business-friendly, while availability in the Gemini Enterprise Agent Platform Model Garden is still forthcoming. The model uses the Gemma 4 architecture and has 740 million parameters.
Why it matters
Engineers can shrink output vector sizes on the fly, from 768 dimensions to 512, 256, or 128, and the announcement claims this cuts storage by up to 6x. Its 8K-token context window is four times the size of the one in EmbeddingGemma 1.
Read the original at Google DeepMind