This survey explores the capabilities of Large Language Models (LLMs) in solving puzzles, highlighting their potential and challenges in complex reasoning tasks. The paper introduces a taxonomy of puzzles into rule-based and rule-less categories, assessing LLMs through various methodologies such as prompting techniques, neuro-symbolic approaches, and fine-tuning. It reviews existing datasets and benchmarks, identifying significant challenges in complex puzzle scenarios. The survey emphasizes the disparity between LLM capabilities and human-like reasoning, particularly in tasks requiring advanced logical inference. It also discusses the need for novel strategies and richer datasets to improve LLMs' puzzle-solving proficiency. The paper categorizes puzzles into rule-based (deterministic and stochastic) and rule-less (riddles, programming, and commonsense reasoning). Rule-based puzzles require logical deduction and strategic foresight, while rule-less puzzles rely on flexible thinking and real-world knowledge. The survey analyzes various methods for solving puzzles, including prompting techniques, neuro-symbolic approaches, and fine-tuning. It evaluates the effectiveness of these methods on different puzzle types, highlighting the challenges in solving stochastic and programming puzzles. The paper also discusses datasets, benchmarks, and tasks used to evaluate LLMs in puzzle-solving. It highlights the versatility of LLMs in solving rule-based puzzles and the challenges in solving rule-less puzzles. The survey identifies current challenges in puzzle-solving, such as the difficulty of solving stochastic puzzles and the limitations of LLMs in programming and commonsense reasoning tasks. It also discusses the impact of question format on puzzle-solving effectiveness, with multiple-choice formats being easier for LLMs than free-text formats. The paper concludes that while LLMs show promise in puzzle-solving, there are still significant challenges to overcome. It emphasizes the need for advanced methodologies and diverse datasets to enhance LLMs' proficiency in puzzle-solving. The survey also highlights the importance of interactive and conversational reasoning in commonsense understanding and the potential of neuro-symbolic techniques in translating natural language into code for puzzle-solving. Overall, the survey provides a comprehensive overview of the current state of puzzle-solving using LLMs and identifies areas for future research.This survey explores the capabilities of Large Language Models (LLMs) in solving puzzles, highlighting their potential and challenges in complex reasoning tasks. The paper introduces a taxonomy of puzzles into rule-based and rule-less categories, assessing LLMs through various methodologies such as prompting techniques, neuro-symbolic approaches, and fine-tuning. It reviews existing datasets and benchmarks, identifying significant challenges in complex puzzle scenarios. The survey emphasizes the disparity between LLM capabilities and human-like reasoning, particularly in tasks requiring advanced logical inference. It also discusses the need for novel strategies and richer datasets to improve LLMs' puzzle-solving proficiency. The paper categorizes puzzles into rule-based (deterministic and stochastic) and rule-less (riddles, programming, and commonsense reasoning). Rule-based puzzles require logical deduction and strategic foresight, while rule-less puzzles rely on flexible thinking and real-world knowledge. The survey analyzes various methods for solving puzzles, including prompting techniques, neuro-symbolic approaches, and fine-tuning. It evaluates the effectiveness of these methods on different puzzle types, highlighting the challenges in solving stochastic and programming puzzles. The paper also discusses datasets, benchmarks, and tasks used to evaluate LLMs in puzzle-solving. It highlights the versatility of LLMs in solving rule-based puzzles and the challenges in solving rule-less puzzles. The survey identifies current challenges in puzzle-solving, such as the difficulty of solving stochastic puzzles and the limitations of LLMs in programming and commonsense reasoning tasks. It also discusses the impact of question format on puzzle-solving effectiveness, with multiple-choice formats being easier for LLMs than free-text formats. The paper concludes that while LLMs show promise in puzzle-solving, there are still significant challenges to overcome. It emphasizes the need for advanced methodologies and diverse datasets to enhance LLMs' proficiency in puzzle-solving. The survey also highlights the importance of interactive and conversational reasoning in commonsense understanding and the potential of neuro-symbolic techniques in translating natural language into code for puzzle-solving. Overall, the survey provides a comprehensive overview of the current state of puzzle-solving using LLMs and identifies areas for future research.