Q-Bench-Video: leaderboard

Each model answers 1,186 test-set questions about the visual quality of videos and video pairs (technical, aesthetic, temporal and AI-generated distortions), as yes-or-no, multiple-choice or open-ended answers; the score is overall accuracy, with open answers graded by GPT-4o. Q-Bench asks the same about still images.

Metric: Overall accuracy (%). Source: raw.githubusercontent.com. Saturation forecast: Estimated already saturated. 10 models tracked. Updated 17 Nov 2026.

Top models

#ModelScoreOverall rank
1GPT-4 Turbo56.36#1008
2mPLUG-Owl3-7B52.39#2811
3GPT-4o Mini (2024-07-18)52.2#1494
4InternVL-Chat-ViT-6B-Vicuna-7B51.11
5VILA-1.5-8B49.41#5342
6llava-v1.6-mistral-7B47.52#3652
7LLaVA-v1.5-13B45.64#7696
8Video-LLaVA-7B43.49#3875
9mPLUG-Owl243.43#8207
10ST-LLM35.42#6018

No result here: #1 Claude Fable 5.2, #2 GPT-6.2 Astra, #3 Gemini 4 Carbon.

Interactive version: theaggregate.ai/benchmark?slug=q-bench-video · How It Works · Data refreshed daily, snapshot 2026-11-17.