Google Ships EmbeddingGemma 2: Open Multimodal Embeddings in One Vector
_philschmid · x · 2026-10-07
EmbeddingGemma 2 is out — Google's first natively multimodal embedding model, built on Gemma 4 and released under Apache 2.0. It embeds 100+ languages, code, images, audio, and video into a single vector space with 8,192 context, in four sizes (740M omni, 440M text+vision, 570M text+audio, 270M text-only), posts a +14% jump on MTEB Code, and supports Matryoshka dimensions (768/512/256/128). Available in Sentence Transformers and LiteRT-LM.
More from Multimodal
- First-try image comparison: GPT edges out Nano Banana 2.1 on detail — techhalla · 2026-10-07
- Photo round: GPT Image 2.5 edges Nano Banana 2.1 on detail in 20-prompt test — techhalla · 2026-10-07
- Composition round: Nano Banana 2.1 and GPT Image 2.5 trade blows — techhalla · 2026-10-07
- Nano Banana 2.1 vs GPT Image 2.5: 20-prompt test scored on adherence, aesthetics, composition — techhalla · 2026-10-07
- Six hours after launch: is Nano Banana 2.1 better than GPT Image 2.5? — techhalla · 2026-10-07
- Seedance 2.0 demo: 'The wrong side of the Cathedral' generated via OpenArt — piotrbinkowski · 2026-10-07