Reproduce “VAHBench: A Harness Benchmark for Agentic Long-Video Generation” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Hongbo Zhang, Liuyang Song, Hang Zhou, Xinchen Liu, Wei He, Kai Han, Yunhe Wang

VAHBench evaluates whether video agents deliver requested series by scoring genuine footage completeness separately from basic video quality and execution-trace reliability. Across 240 runs on five English briefs, Codex’s system mean had higher quality than Claude Code’s (0.616 vs. 0.513) but lower completeness (0.136 vs. 0.814). The benchmark measures basic defects, not aesthetics, and its brief variations limit causal interpretation.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.