List your company
ResearchFormerly WizWand

Video Reasoning on VSI-Bench (test)

38.6AccuracyVideo-ToC
27.26430.20733.1536.093Nov 14, 2025Dec 10, 2025Jan 6, 2026Feb 1, 2026Feb 28, 2026Mar 26, 2026Apr 22, 2026
Updated 6mo ago

Evaluation results

MethodLinks
Video-ToCFrames=642026.0438.6
Video-ToC-SFTFrames=642026.0437.6
Video-R1Frames=642026.0437.1
VideoRFTBase Model=Qwen2.5-VL-7B, Training Strategy=RFT2025.1136.8
VIDEOP2RBase Model=Qwen2.5-VL-7B, Training Strategy=Process-aware Reinforcement Learning2025.1136.8
Video-ToCFrames=322026.0436.4
Video-R1Base Model=Qwen2.5-VL-7B, Training Strategy=RFT2025.1135.8
Video-ToC-SFTFrames=322026.0435.8
Video-R1Frames=322026.0435.8
Video-ToCFrames=162026.0435.3
Video-ToC-SFTFrames=162026.0434.8
Video-R1-SFTFrames=642026.0434.8
Video-R1Frames=162026.0434.6
GPT-4oFrames=-2026.0434
VideoChat-R1Base Model=Qwen2.5-VL-7B, Training Strategy=RFT2025.1133.9
VersaVid-R1Base Model=Qwen2.5-VL-7B, Training Strategy=RFT2025.1133.7
Video-R1-SFTFrames=322026.0433.3
LLaVA-OneVision-7BModel Scale=7B, Training Strategy=Open-Source2025.1132.4
LLaVA-OneVision-7BFrames=-2026.0432.4
Video-R1-SFTFrames=162026.0431.8
Baseline (Qwen2.5-VL-7B)Frames=642026.0431.4
Qwen2.5-VL-7BModel Scale=7B, Training Strategy=Open-Source2025.1130.1
Baseline (Qwen2.5-VL-7B)Frames=322026.0430.1
LongVA-7BModel Scale=7B, Training Strategy=Open-Source2025.1129.2
LongVA-7BFrames=-2026.0429.2
Time-R1Base Model=Qwen2.5-VL-7B, Training Strategy=RFT2025.1129
VILA-1.5-8BFrames=-2026.0428.9
VideoChat-R1Frames=162026.0428.9
Baseline (Qwen2.5-VL-7B)Frames=162026.0427.7