RULER-Bench: Probing Rule-based Reasoning Abilities of Next-level Video Generation Models for Vision Foundation Intelligence

arXiv 2025 32.7 benchmark

TLDR

RULER-Bench evaluates rule-based reasoning in video generation models across 40 tasks, finding state-of-the-art models achieve only 48.87% rule coherence.

Reasoning

The paper introduces a novel benchmark with fine-grained cognitive rule categories and a GPT-o3-based evaluation protocol that aligns 85% with human judgments. However, its scope is limited to video generation reasoning and the abstract does not provide evidence of broader world-model capabilities or downstream applicability.

Read-first score

Read-first score 32.7, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 5.

Recency 8%

86.7

Uses a gentle age decay so recent papers surface without erasing older foundations. 2025

Methodology quality 25%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=benchmark,evaluation,experiment,metric

Reproducibility 25%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=none

Topical relevance 42%

7.1

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Field roles

Frontier

Rank sensitivity

Stability: volatile; rank range: 406.

Keyword Scores

video world model

world model

world simulator

generative world model

interactive world model

world dynamics prediction

model-based reinforcement learning world model

Resources

Paper