Pause and Think: A Dataset and Benchmark for Video-Grounded Assistive Action Suggestion
For researchers and practitioners in vision-language models, this work demonstrates that targeted reasoning supervision enables compact models to achieve strong performance on grounded video reasoning tasks without large-scale model expansion.
The authors introduce a reasoning-centric training dataset and benchmark for video-grounded assistive action suggestion. Their compact 4B-parameter model achieves 58.0% accuracy, matching GPT-5.2 on scene understanding and surpassing GPT-4o, while using 59x fewer parameters than Qwen3-VL-235B.
Recent Vision-Language Models (VLMs) struggle with grounded reasoning, temporal consistency, and context aware planning in videos. We introduce pause-and-think-T, a reasoning-centric training dataset that encourages models to pause, reason over visual evidence, and produce concise, actionable responses. The dataset promotes structured reasoning prior to answer generation, guiding models toward human-like, scene-grounded assistance. We fine-tune a compact 4B-parameter model and evaluate it on our pause-and-think-B benchmark targeting contextual understanding and goal planning tasks. The model achieves 58.0% accuracy at 59x fewer parameters than Qwen3-VL-235B (58.9%), matching GPT-5.2 on scene understanding and surpassing GPT-4o. Beyond our benchmark, it also shows strong out-of-distribution performance on EgoThink and TempCompass, with substantial gains in affordance, assistance, attribution recognition, situated reasoning, and temporal order, without benchmark-specific training. Our results indicate that targeted reasoning supervision enables compact models to deliver actionable, visually grounded guidance while generalizing beyond training data, without requiring large-scale model expansion.