Meta's new Video Understanding Multimodal Model used Qwen model for training

but check mavi out here, it is by far the best at understanding hours long videos. https://openinterx.com