Jina Embeddings v5 Omni Small by Jina AI — 1024D

1.74B-parameter multimodal omni embedding model accepting text, images, video, and audio with shared vector space aligned to text-only embeddings. Supports 4 tasks (retrieval, classification, clustering, text-matching) with task-specific adapters. 1024-dim embeddings with Matryoshka truncation down to 32 dims. Built on Qwen3 architecture with frozen-tower composition.