Benchmark RadarAI BENCHMARK PROFILE
VGI-BENCH
Contains 27 tasks and 810 instances organized by task domains and skill tags to evaluate visual reasoning capabilities of video generation models.
- Released
- 2026-08-20
- Readiness
- Paper only
- Primary field
- General AI
Why it matters
Provides fine-grained evaluation of video generation models' visual reasoning, addressing input alignment, process validity, and task difficulty calibration.
Motivation
Recent studies suggest that video generation models can exhibit certain forms of zero-shot visual reasoning through generated frames.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.