Trade-offs in Large Reasoning Models: An Empirical Analysis of Deliberative and Adaptive Reasoning over Foundational Capabilities
Weixiang Zhao, Xingyu Sui, Jiahe Guo, Yulin Hu, Yang Deng, Yanyan Zhao, Xuda Zhi, Yongbo Huang, Hao He, Wanxiang Che, Ting Liu, Bing Qin
TL;DR
This work investigates the trade-offs of equipping large reasoning models with deliberative reasoning through distillation or reinforcement learning. Across DeepSeek, Qwen, and LLaMA families at 7B–32B scales, it shows that stronger deliberative reasoning markedly degrades foundational capabilities like helpfulness and safety while increasing inference costs. The authors demonstrate that adaptive reasoning modes—Zero-Thinking, Less-Thinking, and Summary-Thinking—can mitigate some of these drawbacks and improve performance on various general tasks. They argue for designing LRMs capable of dynamically allocating inference-time compute according to task characteristics to achieve balanced, versatile performance.
Abstract
Recent advancements in Large Reasoning Models (LRMs), such as OpenAI's o1/o3 and DeepSeek-R1, have demonstrated remarkable performance in specialized reasoning tasks through human-like deliberative thinking and long chain-of-thought reasoning. However, our systematic evaluation across various model families (DeepSeek, Qwen, and LLaMA) and scales (7B to 32B) reveals that acquiring these deliberative reasoning capabilities significantly reduces the foundational capabilities of LRMs, including notable declines in helpfulness and harmlessness, alongside substantially increased inference costs. Importantly, we demonstrate that adaptive reasoning -- employing modes like Zero-Thinking, Less-Thinking, and Summary-Thinking -- can effectively alleviate these drawbacks. Our empirical insights underline the critical need for developing more versatile LRMs capable of dynamically allocating inference-time compute according to specific task characteristics.