MLVU: Benchmarking Multi-task Long Video Understanding

Think of MLVU as a very tough exam designed specifically to see how good AI systems are at watching and really understanding long videos (like full TV episodes, sports matches, or livestreams), not just short clips. It doesn’t build a product itself; it’s a standardized test that tells you which AI models are actually good at following what’s happening over long periods of time and across different types of tasks.