Video-Arena: A Scalable Agentic Video Benchmark for Spatiotemporal Capability
To close this gap, we introduce Video-Arena, a benchmark that turns competitive games into scalable, verifiable tests of how well models understand evolving world states from videos.
ProofPaper ↗
Key points
- Understanding and tracking a dynamic world are fundamental to visual intelligence.
- Existing video benchmarks provide limited insight into whether models can maintain a coherent understanding of evolving world state.
- Specifically, inspired by how humans understand dynamic events, we evaluate spatiotemporal world state understanding and tracking through three task categories: perceiving critical events, interpreting state updates, and tracking the resulting states.
- To keep evaluation affordable and reproducible, we curate a golden set of 180 tasks across 9 games and run evaluations under two protocols: native visual understanding and tool-assisted agentic analysis.
Sources (1)
- [1]Video-Arena: A Scalable Agentic Video Benchmark for Spatiotemporal CapabilityHugging Face Daily Papers · Oct 9, 12:00 AM
To close this gap, we introduce Video-Arena, a benchmark that turns competitive games into scalable, verifiable tests of how well models understand evolving world states from videos.
Understanding and tracking a dynamic world are fundamental to visual intelligence.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 8, 2026REDaCT: Collapse-Free Unlearning in Large Reasoning Models
- Oct 8, 2026openai/codex rust-v0.162.0: 0.162.0
- Oct 8, 2026Show HN: Jevman – AI decision models play Pac-Man
- Oct 8, 2026Microsoft Joins the Local AI Push
- Oct 8, 2026Understanding Jev, the new model everyone is talking about
- Oct 7, 2026[AINews] Claude Haiku 5.5 — better than GPT-6 Luna at the same pricing