MM-BRIGHT: A Multi-Task Multimodal Benchmark for Reasoning-Intensive Retrieval

By Abdelrahman Abdallah, Mohamed Darwish Mounis, Mahmoud Abdalla, M. Kasem, Mostafa Farouk Senussi, Mohamed Mahmoud, Mohammed Ali, Adam Jatowt, Hyun-Soo Kang. 2026

Subfields: Information Retrieval & Search, Multimodal Learning, Evaluation & Benchmarks

Citations in the corpus

• Argus-Retriever: Vision-LLM Late-Interaction Retrieval with Region-Aware Query-Conditioned MoE for Visual Document Retrieval (2026, unknown): “Early multimodal retrievers followed the Con-trastive Language–Image Pre-training (CLIP) recipe, encoding an image and a text query into global vectors and comparing them with a single similarity score (Radford et al., 2021; Abdallah et al., 2026b).”
• BRIDGE: Multimodal-to-Text Retrieval via Reinforcement-Learned Query Alignment (2026, unknown): “We evaluate BRIDGE on MM-BRIGHT [6], the first reasoning-intensive multimodal retrieval benchmark.”
• MARVEL: Multimodal Adaptive Reasoning-intensiVe Expand-rerank and retrievaL (2026, unknown): “MM-BRIGHT [5] confirmed this directly — the best multimodal encoder (27.6 nDCG@10) underperforms strong text-only retrievers, revealing that visual encoding capacity is not the bottleneck.”
• HIVE: Query, Hypothesize, Verify An LLM Framework for Multimodal Reasoning-Intensive Retrieval (2026, unknown): “We evaluate HIVE on MM-BRIGHT [6], the first reasoning-intensive multimodal retrieval benchmark.”