New benchmark tests vision-language models' spatiotemporal reasoning via Atari gameplay
Researchers introduced AtariBench, a benchmark on arXiv that evaluates how well vision-language models reason about space and time by playing Atari games, moving beyond static image benchmarks.
Read the full story at arXiv cs.CV ↗
Timeline · 1 report
- 2026-11-09 05:00 · arXiv cs.CV
AtariBench: Evaluating Spatiotemporal Reasoning of Vision-Language Models in Atari Gameplay