Benchmark RadarSVI-Bench
SVI-Bench evaluates vision-language models on strategic video intelligence using sports as a microworld. It includes 9 tasks across 4 pillars: perception, reasoning, simulation, and agency, using basketball, soccer, and hockey videos with annotated actions and reports.
- Released
- 2026-05-29
- Readiness
- Runnable
- Primary field
- General AI
Why it matters
Existing video benchmarks lack verifiable ground truth for causal and strategic reasoning. SVI-Bench combines real-world multi-agent complexity with verifiable rules and outcomes, enabling evaluation of higher-level cognitive capabilities in video understanding.
Motivation
True video intelligence demands more than recognizing what is visible: it requires reasoning about why events unfold, predicting what would change under different conditions, and deciding what to do next.
Primary resources
Benchmark Radar records only publicly supported details and links back to primary sources for verification.