The recent advancements in Vision-Language Models (VLMs) and Large Language Models (LLMs) have shown significant progress in areas such as visual reasoning, mathematical problem-solving, and complex task supervision. VLMs, while demonstrating advanced reasoning capabilities across text and image modalities, still face challenges in understanding and reasoning about elementary visual concepts, indicating a gap between human-like visual reasoning and machine cognition. In mathematical reasoning, the introduction of step-guided reasoning methods has improved accuracy, suggesting a shift towards more reflective and step-by-step approaches in problem-solving. Additionally, the use of 'weak teacher models' for supervision in hard reasoning tasks has revealed that step-wise error rates are critical factors influencing training performance, opening new avenues for data augmentation strategies. Visual Premise Proving (VPP) has been introduced as a novel task to refine chart question answering, emphasizing the importance of integrating reasoning with visual comprehension. Lastly, the VisAidMath benchmark highlights deficiencies in visual-aided mathematical reasoning, particularly in the implicit visual reasoning process, pointing to areas for future research. Overall, these developments indicate a move towards more nuanced and integrated approaches in both visual and mathematical reasoning tasks, with a focus on improving the reflective and step-by-step reasoning capabilities of models.