UEmbed

model
Alibaba

Multimodal embedder (2B/4B/9B, Qwen3.5 backbones) emitting dense + SPLADE-style sparse embeddings in one causal pass — text, image, and video, inverted-index-compatible for hybrid retrieval. CC-BY-4.0.

HuggingFace (9B)Paper (arXiv)GitHub

Model Details

Architecture DENSE
Base model qwen3.5

Paper

arXiv HTML
multimodalopen-weightretrieval