MMSearch-OPD: Easy Made Hard, Hard Made Learnable for Long-Horizon Multimodal Search Agents
We introduce MMSearch-OPD, combining controllable data refinement with joint reinforcement learning (RL) and on-policy distillation (OPD).
ProofPaper ↗
Key points
- Multimodal deep search is moving from short queries to long-horizon investigations that continually acquire evidence.
- Our Refine pipeline performs targeted expansion of reasoning chains derived from existing simple multimodal questions.
- Parameterized transformations control search depth and cross-modal evidence requirements while reusing existing image--question pairs.
- Experiments show gains from data refinement and further improvements from joint training after SFT, indicating stronger long-horizon multimodal deep search capabilities.
Sources (2)
- [1]MMSearch-OPD: Easy Made Hard, Hard Made Learnable for Long-Horizon Multimodal Search AgentsarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 15, 05:16 PM
We introduce MMSearch-OPD, combining controllable data refinement with joint reinforcement learning (RL) and on-policy distillation (OPD).
Multimodal deep search is moving from short queries to long-horizon investigations that continually acquire evidence.
- [2]MMSearch-OPD: Easy Made Hard, Hard Made Learnable for Long-Horizon Multimodal Search AgentsHugging Face Daily Papers · Oct 15, 12:00 AM · same content
Extractive summary: sentences quoted from the sources.
Before this
- Oct 14, 2026RM-OPD: On-Policy Distillation for Generative Reward Models
- Oct 14, 2026Occupancy Modeling via Drifting: One-Step Generation Enables Direct Policy Extraction
- Oct 14, 2026The Design Space of Tri-Modal Masked Diffusion Models
- Oct 14, 2026More Methods, Better Agents? A Reality Check on Self-Distillation in Agentic Reinforcement Learning
- Oct 13, 2026Demystifying On-Policy Distillation: Where Does the Teacher's Advantage Live?
- Oct 13, 2026Run fast, private agents on-device with LFM2.5-24B-A2B
