The research area of visual object counting and vision-language models (VLMs) is experiencing significant advancements, particularly in addressing the challenges of counting stacked 3D objects and enhancing the counting capabilities of VLMs. Innovations in 3D counting methods are focusing on decomposing the task into geometric reconstruction and depth analysis, enabling accurate counting of irregularly stacked objects. Meanwhile, VLMs are being improved through divide-and-conquer strategies that enhance their ability to count objects in diverse datasets without additional training. Additionally, the development of versatile VLMs like PaliGemma 2 is broadening the scope of transfer tasks, including OCR-related tasks and fine-grained captioning, achieving state-of-the-art results. These developments collectively push the boundaries of what is possible in visual counting and VLM applications, making significant strides in both accuracy and versatility.