VideoChat3: Fully Open Video MLLM for Efficient and Generalist Video Understanding

MCG-NJU

Research official 1 src. ~1 min

A fully open (weights, data, and training recipe) 4B-parameter video multimodal LLM using an Inflated 3D Vision Transformer with adaptive frame resolution, paired with a scalable video-data synthesis pipeline producing general, long-form, and streaming-video datasets. Claims to surpass prior open-source models with equal or larger parameter counts across multiple video benchmarks.

Why it matters

24 upvotes on Hugging Face Daily Papers; notable for full openness (data + recipe, not just weights) at a small parameter count that beats larger open models.

Importance: 3/5

HF Daily Paper (24 upvotes); fully-open small video MLLM outperforming larger open models.

Sources