A Multimodal Multilingual Multicultural Benchmark for
LLMs' Contextual and Cultural Knowledge and Thinking

Presented by AVMeme Team 🔔

❮
❯
▼

Testing LLMs Content, Context & Cultural Understanding

Covering well-known, multilingual Internet speeches, songs, music, and sound effects.

Question Types

Sound Categories

Languages

We are interested in whether multimodal LLMs can perceive beyond the media's content, including its
context, how people feel and use it, and where it situates in our shared cultural world.
▼

Excel at Language, but Struggle with Context, Humor, and Usage

All models except Gemini 3 lack Multimedia World Knowledge.

Performance by Question Type

Performance by Sound Category

▼
ArXiv Logo Paper Hugging Face Logo Download

Valhalla of AVMeme

Xilin JiangColumbia Qiaolin WangColumbia Junkai WuUW Xiaomin HeColumbia Zhongweiyang XuUIUC Yinghao MaQMUL Minshuo PiaoColumbia Kaiyi YangUIUC Xiuwen ZhengUIUC Riki ShimizuColumbia Yicong ChenUW Arsalan FirooziColumbia Gavin MischlerColumbia Sukru Samet DindarColumbia Richard AntonelloColumbia Linyang HeColumbia Tsun-An HsiehUIUC Xulin FanUIUC Yulun WuUIUC Yuesheng MaColumbia Chaitanya AmballaUIUC Weixiong ChenQMUL Jiarui HaiJHU Ruisi LiNYU Vishal ChoudhariColumbia Cong HanGoogle Yinghao Aaron LiMeta Adeen FlinkerNYU Mounya ElhilaliJHU Emmanouil BenetosQMUL Mark Hasegawa-JohnsonUIUC Romit Roy ChoudhuryUIUC Nima MesgaraniColumbia
Participating Institutions