WorldOdysseyBench: An Open-World Benchmark for Long-Horizon Stability of Interactive World Models

arXiv 2026 39.9 benchmark

TLDR

Introduces WorldOdysseyBench, a benchmark for long-horizon stability of interactive world models across action, vision, physics, and memory dimensions.

Reasoning

Strengths include a comprehensive multi-dimensional benchmark with novel metrics and diverse real-world scenes; weaknesses are that the abstract does not detail specific model failures or limitations beyond moderate scores.

Read-first score

Read-first score 39.9, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 39.

Recency 6%

100

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

Methodology quality 18%

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

Topical relevance 29%

55.7

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 18%

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

Citation impact 18%

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

Citation velocity 12%

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: 220.

Keyword Scores

interactive world model

world model

world dynamics prediction

world simulator

video world model

generative world model

model-based reinforcement learning world model

Deep Analysis

Innovations

Per-frame action metric that bypasses cross-model semantic scale disparity and exposes failures hidden by trajectory-level evaluation
Segment-based drift metric that captures non-monotonic mid-sequence collapse missed by start-vs-end comparisons
Controllability-gated evaluation over mechanics, optics, and 3D consistency, scoring plausibility under faithful action execution
Action-decoupled memory protocol evaluating scene memory via transition-localized 3D point-cloud reconstruction and subject memory via tracking-plus-VLM reasoning

Methodology

WorldOdysseyBench is a benchmark comprising 600+ test cases across Nature, Urban, and Indoor scenes in first/third-person views with WASD continuous interaction lasting 10-60 seconds. It evaluates interactive world models on four dimensions—Action, Vision, Physics, and Memory—using tailored metrics and protocols, and tests over 10 open- and closed-source models.

Key Results

No evaluated model reliably satisfies all four dimensions; even the best-performing model achieves only moderate scores.

Resources

Paper