AgentVidBench: A Multi-Hop Video QA Benchmark for Testing MLLM Agents

Kangwook_Lee · x · 2026-09-21

Kangwook Lee's team released AgentVidBench on arXiv, a multi-hop video question-answering benchmark evaluating spatial, temporal, and causal reasoning in MLLM agents.

Highlights:

Related event: AgentVidBench: A Multi-Hop Video QA Benchmark for MLLM Agents(2 posts)→

Original post →

More from Multimodal

Multimodal channel →