Physics-IQ Verified

arXiv 2026 67.4 benchmark, method

TLDR

Audits and improves the Physics-IQ benchmark for evaluating physical understanding in video generative models, refining prompts and scoring.

Reasoning

Strengths: Systematic audit with concrete improvements (57.6% sample refinement) and empirical comparison across six models. Weaknesses: Limited to image-to-video models; no new real-world data collection, only benchmark refinement.

Read-first score

Read-first score 67.4, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 35.

Recency 6%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Citation impact 18%

94.3

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. citation_normalized_percentile=0.94337365

Methodology quality 18%

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

Reproducibility 18%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=True; dataset=False; markers=code,github

Topical relevance 29%

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

Citation velocity 12%

Citation velocity estimates citations per publication-year to reduce old-paper bias. velocity=0.00

Field roles

FoundationFrontierBridgeMethodology anchorReproducibility anchor

Rank sensitivity

Stability: volatile; rank range: 460.

Keyword Scores

video world model

world model

generative world model

world dynamics prediction

world simulator

interactive world model

model-based reinforcement learning world model

Deep Analysis

Innovations

Improving prompt and ground-truth quality to reduce the influence of confounding factors
Introducing a sample-level scoring system that weights each sample and metric equally

Methodology

The authors conduct a systematic audit of the Physics-IQ benchmark, exposing shortcomings and proposing three solutions. They refine 57.6% of all samples and improve 34.8% of prompts. A comparison study using six image-to-video generative models is performed to evaluate the refined benchmark.

Key Results

The refined benchmark, Physics-IQ Verified, shows moderate but meaningful ranking changes among six image-to-video models with Kendall's τ = 0.46.

Resources

Paper Code