The field of Large Language Models (LLMs) is rapidly advancing, with a focus on improving their performance, efficiency, and alignment with human preferences. Recent studies have demonstrated the effectiveness of LLMs in various domains, including mathematics and pharmacology, with some models achieving high scores in exams and outperforming others. The development of new training paradigms, such as preference-oriented instruction-tuned reward models, and optimization methods, like intrinsic confidence-driven group relative preference optimization, are addressing key challenges in LLM training, including data efficiency, reward overoptimization, and exploration instability. These advancements have significant implications for the practical deployment of LLMs in real-world applications. Noteworthy papers include: Beyond Surface-Level Similarity, which proposes a hierarchical contamination detection framework for synthetic training data, and Evaluating Large Language Models on the 2026 Korean CSAT Mathematics Exam, which demonstrates the mathematical reasoning capabilities of state-of-the-art LLMs in a contamination-free evaluation environment.