Large Reasoning Models in Agent Scenarios: Exploring the Necessity of Reasoning Capabilities

2025-03-14

AI Summary

This work investigates the necessity and applicability boundaries of Large Reasoning Models (LRMs) in agent-based tasks. Addressing three core questions—whether LRMs outperform Large Language Models (LLMs), whether hybrid architectures enhance performance, and what inference costs and maladaptive behaviors LRMs entail—we introduce LaRMA, a structured evaluation framework comprising nine benchmark tasks. We systematically compare LLMs (e.g., Claude 3.5 Sonnet) and LRMs (e.g., DeepSeek-R1) across tool invocation, planning, and problem solving. We propose, for the first time, an LLM-as-Executor / LRM-as-Reflector collaborative architecture and establish an empirical paradigm for reasoning utility assessment. Results show LRMs improve planning accuracy by 23.6%, whereas LLMs exhibit superior efficiency and stability in tool execution. The hybrid architecture achieves a 17.2% gain in overall performance but also uncovers novel LRM challenges: factual neglect, response latency, and over-reasoning.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Cognitive Modeling & Cognitive Systems: Agent Architectures
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems

Abstract

The rise of Large Reasoning Models (LRMs) signifies a paradigm shift toward advanced computational reasoning. Yet, this progress disrupts traditional agent frameworks, traditionally anchored by execution-oriented Large Language Models (LLMs). To explore this transformation, we propose the LaRMA framework, encompassing nine tasks across Tool Usage, Plan Design, and Problem Solving, assessed with three top LLMs (e.g., Claude3.5-sonnet) and five leading LRMs (e.g., DeepSeek-R1). Our findings address four research questions: LRMs surpass LLMs in reasoning-intensive tasks like Plan Design, leveraging iterative reflection for superior outcomes; LLMs excel in execution-driven tasks such as Tool Usage, prioritizing efficiency; hybrid LLM-LRM configurations, pairing LLMs as actors with LRMs as reflectors, optimize agent performance by blending execution speed with reasoning depth; and LRMs' enhanced reasoning incurs higher computational costs, prolonged processing, and behavioral challenges, including overthinking and fact-ignoring tendencies. This study fosters deeper inquiry into LRMs' balance of deep thinking and overthinking, laying a critical foundation for future agent design advancements.

Problem

• Explores necessity of reasoning capabilities in agent scenarios.
• Compares performance of LRMs and LLMs in reasoning tasks.
• Investigates hybrid configurations for optimizing agent performance.

Innovation

• LaRMA framework integrates nine reasoning tasks
• Hybrid LLM-LRM optimizes agent performance
• LRMs enhance reasoning but increase computational costs