Benchmark RadarAI BENCHMARK PROFILE
MemoBench
MemoBench evaluates world modeling in video generation via a disappear-and-reappear paradigm. It includes 360 ground-truth clips and an evaluation suite with automated metrics and VQA across four diagnostic pillars.
- Released
- 2026-06-25
- Readiness
- Runnable
- Primary field
- General AI
Why it matters
Current benchmarks assess memory consistency only when objects remain visible; MemoBench tests object recovery after occlusion in dynamic scenes, providing a reusable diagnostic for world models.
Motivation
Video generation models aspire to simulate dynamic environments, and several benchmarks now evaluate memory consistency across frames.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.