ViMU: leaderboard

ViMU (2026) benchmarks video metaphorical understanding: whether models grasp the implicit, ironic or socially coded meaning of short online videos. The score is the average in percent of its four tasks: open-ended interpretation, evidence grounding, and identification of rhetoric mechanisms and social value signals; higher is better.

Metric: Average (%) of the four task scores (open-ended interpretation, evidence grounding, rhetoric mechanism and social value signal identification) on ViMU, short online videos whose meaning lies in subtext (irony, mockery, criticism), uniformly sampled frames, zero-shot through official implementations or APIs; questions and references written by GPT-5.4 and reviewed by human experts; higher is better. Source: arxiv.org. Saturation forecast: Around July 2028. 16 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1O4 Mini46.91#368
2Grok 4.1 Fast46.28#427
3Qwen 3.5 27B45.91#375
4GPT-5.244.67#185
5Gemini 3 Flash (Preview)44.31#152
6Qwen 3 VL 32B Instruct41.64#606
7Seed 2.0 Lite40.62#300
8MiMo-V2-Omni38.14#465
9GPT-5.4 Mini36.64#406
10Gemma 3 27B (IT)36.43#1324
11Ministral 3 14B35.45#1870
12Ministral 8B34.79#4857
13GLM-4.5V25.94#754
14Gemma 3 4B (IT)23.28#4139
15Claude 3 Haiku22.9#2663

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

Interactive version: theaggregate.ai/benchmark?slug=vimu · How It Works · Data refreshed daily, snapshot 2026-11-21.