LVBench is a benchmark specifically designed for long video understanding, aimed at advancing the ability of multimodal large language models to comprehend videos spanning several hours. This is crucial for real-world applications such as long-term decision-making, in-depth film reviews and discussions, and live sports commentary.