A general VLM as a multimodal reranker: single-pass choices match dedicated rerankers
Zhuchenyang Liu · hf · 2026-10-21
A new paper frames multimodal reranking as a System One decision: a general vision-language model performs reranking via single-pass multiple-choice questions instead of a dedicated reranker.
On MMEB-V2 and ViDoRe V3, a zero-shot 27B VLM matches or exceeds dedicated 2B/8B rerankers, with lightweight decision tuning pushing performance further—collapsing the retrieve-rerank-generate pipeline into one model and eliminating reranker training/serving costs.
More from Multimodal
- OmniCam from Tencent Hunyuan generates camera trajectories from panoramas and text — Zhenyang Liu · 2026-10-21
- Reddit user shares AI-made Death Note short film for TheLoopHole Project — neptune765 · 2026-10-21
- Hands-on with Fibo 2.0 image editing: solid 360° turns, but object occlusion fails — Big_Parsnip_9053 · 2026-10-21
- Minimax H3 Character Replacement: Ref2va 8-Step LoRA Transfers Facial Emotion Acting — mexxmillion · 2026-10-21
- Painting Change Map guides fine-grained video generation of painting processes — MHUGLab · 2026-10-21
- Open-source Hermes video toolchain turns agent requests into finished MP4s — Teknium · 2026-10-21