Research
VibeVideoBench benchmarks agents that write code to generate videos
A 239-task benchmark in English and Chinese tests agents that render MP4 videos with code and no video generation model. GPT-6 Astra and Claude Opus 5.5 rank first and second, and the top system satisfies every core requirement on 76.3% of tasks.
Every outlet that ran it 1 article
Hugging Face PapersResearchLead
VibeVideoBench: Evaluating Multimodal Agents on Code-Driven Video Generation