Back to Project
V

VibeVideoBench

Project Oct 28, 2026

A research benchmark testing whether complete coding-agent systems can deliver rendered videos from natural-language briefs within one hour.

Visit primary link 3 sources attached
English VibeVideoBench diagram shows video tasks, rubric construction and a one-hour agent rollout with delivery and quality evaluation.
Author-published benchmark design diagram from Figure 2. The evaluation setup and findings are the research team's, not independently reproduced by AITrending. Source: Muxi Diao et al. / VibeVideoBench
agent-ecosystemmultimodalverification

What happened lately

VibeVideoBench timeline

VibeVideoBench evaluates nine complete agent-and-harness systems on 239 video-production tasks across 12 categories. Each run receives a fresh Docker sandbox, eight CPUs, 32 GB of memory, no GPU and a 60-minute deadline. The evaluation asks for a playable final video, rather than treating code generation alone as delivery.

The authors report results for 2,151 rollouts using programmatic delivery checks, task-specific rubrics and quality judging. The paper first appeared on October 26, 2026 and was revised on October 27. At review, the project’s code and dataset links were marked coming soon. Its comparisons describe these particular systems and conditions, not universal model capabilities or independently reproduced AITrending results.