Selection-Based Structured Reasoning: Toward Efficient Multimodal Search Agents

Researchers introduced Selection-Based Structured Reasoning, a method that replaces free-form reasoning with selection-based structured reasoning for multimodal search agents, targeting small models whose limited capacity produces lengthy reasoning with little useful guidance for action generation at substantial inference cost. The approach aims to cut inference cost while improving action generation for small multimodal agents.

Multimodal agents commonly generate free-form reasoning before each action. For small models, limited model capacity can result in lengthy reasoning that provides little useful guidance for action generation while incurring substantial inference cost. To address this challenge, we introduce Selectio