REST: A Stress-Testing Framework for Evaluating Multi-Problem Reasoning in Large Reasoning Models

The development of REST (Reasoning Evaluation through Simultaneous Testing) introduces a significant advancement in the assessment of Large Reasoning Models (LRMs) by shifting from traditional single-question benchmarks to a multi-problem stress-testing framework. This approach aims to better evaluate LRMs' multi-context reasoning abilities, which are more representative of real-world problem-solving scenarios, addressing limitations of existing benchmarks like GSM8K and MATH that focus on isolated questions and lead to saturated performance scores. By challenging LRMs with simultaneous, multi-problem tasks, REST enhances the discriminative power of evaluations, enabling more accurate differentiation