ReasonBENCH: Benchmarking the (In)Stability of LLM Reasoning

The article introduces ReasonBENCH, a benchmark designed to evaluate the consistency and reliability of Large Language Models (LLMs) in reasoning tasks. The focus on stability suggests an investigation into how LLMs perform across multiple runs or under varying conditions, which is crucial for real-world applications. The use of 'In' in the title hints at the potential for instability, indicating a critical assessment of LLM reasoning capabilities.

• •ReasonBENCH is a benchmark for evaluating LLM reasoning.
• •The benchmark focuses on the stability of LLM reasoning.
• •The research likely investigates the consistency of LLM performance.