PlaySuite: A Large-Scale Benchmark for Interactive Visual Intelligence

2026-10-05

AI Summary

This study addresses the lack of long-term interaction evaluation for multimodal models in dynamic environments by constructing an interactive visual intelligence benchmark based on 5,000 open-source games. Methodologically, it proposes a unified evaluation framework for heterogeneous games alongside a closed-loop interaction system, and introduces a Video-LLM-as-a-Judge mechanism that effectively mitigates data leakage while quantifying game milestones. The research reveals a significant perception-action gap in current models, demonstrating their persistent limitations in spatial localization, action execution, and self-correction. Ultimately, this work establishes a novel paradigm for evaluating embodied intelligence.

Abstract

Recent advances in multimodal foundation models yield strong performance on static perception and reasoning benchmarks, yet such evaluations largely overlook a central aspect of intelligence: acting competently in dynamic environments over extended time horizons. We introduce PlaySuite, a large-scale benchmark for evaluating interactive visual intelligence across more than 5K open-source video games curated from PyWeek and itch.io. Spanning diverse genres and engines, including Pygame, HTML5, Godot, and Unity, these independent games are largely out-of-distribution for current models, reducing the likelihood that success can be achieved by retrieving memorized walkthroughs or web-scale training artifacts. To enable scalable evaluation across heterogeneous titles, we develop a unified closed-loop interaction framework optimized for HPC clusters alongside a Video-LLM-as-a-judge protocol that maps observable gameplay milestones to standardized progress levels. We evaluate fourteen recent open models spanning vision-language models, computer-use agents, and vision-language-action models. Our results yield strong evidence of a perception-action gap: despite strong reasoning capabilities, current models struggle to make sustained progress and exhibit recurring failures in spatial grounding, action execution, and self-correction. PlaySuite provides a reproducible and extensible testbed for measuring progress from visual perception to goal-directed interaction, and a foundation for developing models that can act, adapt, and generalize in dynamic visual environments.

Problem

• interactive visual intelligence
• multimodal foundation models
• perception-action gap
• dynamic environments
• benchmark evaluation

Innovation

• Interactive Visual Intelligence
• Large-Scale Benchmark
• Closed-loop Interaction Framework
• Video-LLM-as-a-Judge
• Perception-Action Gap