Google unveils Gemini Embedding 2 multimodal model
Google has launched Gemini Embedding 2, a multimodal embedding model that supports text, images, video, audio, and documents. The model works with more than 100 languages and can process up to 8,192 text tokens, six images, 120-second video clips, and six-page PDFs, employing Matryoshka Representation Learning for flexible output dimensions. Early users have reported strong results on retrieval-augmented generation and semantic search applications.
Read Full Article