Workshop · Accepted ·
Your Model Diversity, Not Method, Determines Reasoning Strategy
Moulik Choraria, Argyrios Gerogiannis, Anirban Das, Supriyo Chakraborty, Berkcan Kapusuzoglu, Chia-Hsuan Lee, Kartik Balasubramaniam, Shi-Xiong Zhang, Sambit Sahu
ICLR 2026 Workshop on Logical Reasoning of Large Language Models · Workshop
Summary
This work argues that reasoning strategies should account for a model's diversity profile and formalizes when depth refinement can outperform parallel sampling.
Research contribution
The paper develops a framework connecting model diversity to the choice between breadth-oriented exploration and depth-oriented refinement.