In the quest to evaluate large language models (LLMs) fairly, researchers have introduced a standardized approach to prompting these systems. By maintaining consistent prompts across multiple models, the aim is to direct them in delivering step-by-step reasoning, referred to as Chain of Thought (COT). The study highlights the strengths and challenges of using this method by also testing a specially engineered prompt for certain hallmark models. The findings suggest that while prompt engineering can enhance model performance marginally, it cannot overcome inherent limitations. The report sheds light on the intriguing observation regarding specific performance drops in GPT4o, hypothesizing its optimization for particular industrial tasks like mathematics and coding. Despite excelling in these areas, the model reportedly struggles with broader reasoning challenges. This phenomenon opens a discussion about the trade-offs associated with fine-tuning AI models for specialized tasks versus their general understanding capabilities, posing questions about balancing application-specific optimization with holistic performance. For those keen on exploring the intricacies of this research, the technical report provides a comprehensive analysis of the methods and findings. It delves into the implications of these standardized assessments and investigates whether foundational AI designs can or should adapt to industry-specific needs. The broader implications of these insights suggest a future where AI specialists might negotiate between performance enhancements and maintaining wide-ranging reasoning abilities.