VideoAuto-R1: Video Auto Reasoning via Thinking Once, Answering Twice

Open paper
Benchmark Model Rank Results
video-question-answering-on-videommmuVideoAuto-R1 (Qwen3-VL-8B)#5Accuracy: 65.0
video-question-answering-on-mmvuVideoAuto-R1 (Qwen3-VL-8B)#12Accuracy: 71.1
mathematical-reasoning-on-mathverseVideoAuto-R1 (Qwen2.5-VL-7B)#20Accuracy: 46.9
zero-shot-video-question-answer-on-video-mme-w-o-subsVideoAuto-R1 (Qwen3-VL-8B)#21Accuracy (%): 71.7
video-question-answering-on-mvbenchVideoAuto-R1 (Qwen3-VL-8B)#22Avg.: 72.0
visual-question-answering-on-mm-vetVideoAuto-R1 (Qwen2.5-VL-7B)#39GPT-4 score: 61.9
multimodal-reasoning-on-mmmu-proVideoAuto-R1 (Qwen2.5-VL-7B)#49Accuracy: 39.8
multimodal-reasoning-on-mathvisionVideoAuto-R1 (Qwen2.5-VL-7B)#51Accuracy: 29.6
multimodal-reasoning-on-mathvistaVideoAuto-R1 (Qwen2.5-VL-7B)#54Accuracy: 73.7
moment-retrieval-on-charades-staVideoAuto-R1 (Qwen3-VL-8B)#74mIoU: 63.7
multimodal-reasoning-on-mmmuVideoAuto-R1 (Qwen2.5-VL-7B)#83Accuracy: 53.8

Rank counts only results with a code link.