Compression then Matching: An Efficient Pre-training Paradigm for Multimodal Embedding Paper • 2511.08480 • Published Nov 11, 2025 • 1