This research explores the potential of iterative self improvement strategies to enhance the performance of Large Language Models (LLMs), both open-source and closed-source. While acknowledging the progress of commercially controlled models like GPT-4o and Gemini 2.0 Flash, this study investigates how iterative refinement can significantly impact LLM capabilities, regardless of size. The core idea is that by enabling LLMs to learn and improve through iterative refinement, they can achieve performance comparable to more advanced systems. The paper proposes a methodology inspired by human problem solving, where models generate outputs, critique them, and refine iteratively for n times. Applicable across various LLMs and tasks, this approach leverages self generated feedback using iterative strategies, Operating in three key stages: initial generation, self-critique, and iterations. Furthermore the framework was tested on more than 14 LLMs, including GPT-4o, Deepseek-R1-Qwen/Llama. The results indicate that the method is scalable and broadly applicable. In smaller LLMs, iterative refinement effectively narrowed the performance gap with larger, resource intensive systems, improving capabilities in complex reasoning, code generation, and multimodal understanding. In general, this work envisions iterative refinement as a priority over mere model scaling, suggesting exploration of hybrid architectures, iterative refinement & energy efficient critics for balanced performance and efficiency.