HappyWorld-Bench

arXiv 2026 44.6 benchmark

TLDR

Introduces HappyWorld-Bench, a benchmark with three tracks and 1,138 video prompts, 300 spatial scenes, and 254 embodied cases to evaluate world models' generation, interaction, and consistency.

Reasoning

The paper provides a comprehensive evaluation framework with hierarchical capabilities, multiple tracks, human A/B comparisons, and automated metrics across many models. Weaknesses are not visible from abstract alone, but the benchmark's reliance on human comparisons may limit scalability.

Read-first score

Read-first score 44.6, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 58.

Recency 5%

100

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

Methodology quality 15%

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

Topical relevance 25%

82.9

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

Reproducibility 15%

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

Graph prestige 15%

5.7

PageRank-style prestige over the local citation graph built from available OpenAlex references. local_pagerank=0.0016

Citation impact 15%

Uses OpenAlex-shaped citation metadata as a bibliometric attention signal, separate from paper quality. cited_by_count=0

Citation velocity 10%

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

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 681.

Keyword Scores

world model

video world model

generative world model

interactive world model

world simulator

world dynamics prediction

model-based reinforcement learning world model

Deep Analysis

Innovations

Introduces HappyWorld-Bench, a comprehensive benchmark for evaluating whether generated worlds remain reliable under exploration, interaction, and modification.
Proposes a hierarchical capability framework of six world capabilities (W1-W6), from generative construction to unified world modeling.
Instantiates the framework across three independent evaluation tracks: video world models, spatial world models, and embodied world models.
Builds and operates HappyWorld-Arena to organize human A/B comparisons and derive model-level Elo ratings, complemented by newly designed automated metrics for behavioral correctness.

Methodology

HappyWorld-Bench comprises 1,138 video prompts, 300 spatial scenes, and 254 embodied test cases organized under the W1-W6 hierarchical framework. It evaluates 14 video world models, 9 spatial systems, and 8 embodied candidates across three tracks. Human A/B comparisons in HappyWorld-Arena produce Elo ratings, while newly designed automated metrics capture behavioral correctness.

Key Results

Video world models show reduced consistency during extended rollouts and revisits; spatial models achieve at best 70.14% placement accuracy and 73.33% edit execution; embodied models struggle to preserve state across multi-step actions and respond precisely to altered action conditions and physical rules.

Limitations

No explicit limitations of the benchmark or study itself are stated in the abstract.

Resources

Paper