The field of multimodal video understanding is rapidly advancing, driven by the development of new benchmarks and evaluation methods. Researchers are focusing on creating more realistic and challenging benchmarks to test the capabilities of large multimodal models. One notable trend is the shift towards open-ended questions and more fine-grained annotations, which require a deeper understanding of the video content. Another area of research is the development of benchmarks for specific tasks, such as cinematographic technique understanding and generation. These efforts aim to push the boundaries of video understanding and generation, enabling more sophisticated applications in film production, appreciation, and beyond. Noteworthy papers in this area include LoVR, which introduces a benchmark for long video-text retrieval, and VideoEval-Pro, which proposes a realistic LVU benchmark with open-ended questions. CineTechBench is also notable for its benchmark on cinematographic technique understanding and generation.