Monitor this situation.
Unsubscribe anytime.
[SITUATION] · [QUIET] · [TECHNOLOGY]
2 clusters · 8 sources · 8 days · First seen · Last updated
Google release of EmbeddingGemma 2 multimodal model
Overview
Google has released EmbeddingGemma 2, an open-weights multimodal embedding AI model based on the Gemma 4 architecture. The model is designed to integrate text, code, images, video, and audio into a single 768-dimensional vector space, facilitating cross-modal retrieval without prior captioning or transcription.
Released under the Apache 2.0 license, the model is optimized for efficient on-device deployment. It features a modular architecture that allows developers to load specific encoders to manage resource usage; for example, a text-only version requires approximately 191MB of RAM, while the full multimodal model requires about 567MB of active RAM on a Pixel 11 Pro.
Technical capabilities include an expanded context window of 8,192 tokens, enabling the processing of up to 5.5 minutes of audio or roughly 58 video frames. Researchers have demonstrated the model’s versatility in speech-to-text tasks, achieving transcription speeds of approximately one hour of audio in 12.6 seconds on an M3 MacBook Air, despite a current word error rate of around 5%.
Entities
Google · EmbeddingGemma 2 · Pixel 11 Pro · Google DeepMind · Gemma 4
Timeline
-
[TECHNOLOGY] 2 sourcesGoogle DeepMind releases EmbeddingGemma 2 multimodal model
Google DeepMind released EmbeddingGemma 2, a modular multimodal embedding model designed for on-device use across text, image, video, and audio, featuring efficient speech-to-text capabilities.
-
[TECHNOLOGY] 6 sourcesGoogle releases EmbeddingGemma 2 multimodal AI model
Google released EmbeddingGemma 2, an open-weights multimodal AI model that integrates text, image, video, and audio into a single vector space for efficient on-device retrieval and search.
Sources
deepmind.google · hackernews.com · it-boltwise.de · macotakara.jp · news.cnyes.com · thebridge.jp · thenewstack.io · weel.co.jp