VCR-Bench: leaderboard

Each model answers all 1,034 questions about 859 videos with step-by-step reasoning, and the score is the share of final answers GPT-4o grades correct (temporal localisation answers by IoU). Its sibling CoT Score board grades the reasoning steps rather than the final answer.

Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 9 models tracked. Updated 17 Nov 2026.

Top models

#ModelScoreOverall rank
1Aria38.2#2189
2Qwen 2.5 VL 72B Instruct37.9#849
3LLaVA-Video-72B-Qwen236.6#1344
4InternVideo2.5-Chat-8B33#2031
5VideoLLaMA3-7B32.5#2015
6InternVL2.5-78B30.9#1036
7Qwen 2.5 VL 7B Instruct30.4#1893
8MiniCPM-o-2.626.9#2417
9InternVL2.5-8B23.9#2401

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

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