WorldAttention: An Efficient Attention Architecture for Interactive Video World Models

arXiv 2026 38.7 method

TLDR

WorldAttention introduces sparse attention and hierarchical KV cache management to enable efficient long-context interactive video world models with low latency.

Reasoning

The paper proposes a system-oriented attention architecture combining Hybrid Sparse Attention and Hierarchical KV Cache to overcome context and memory limitations in interactive video world models. Strengths include a clear problem definition and co-design approach, but the abstract is truncated before results, so quantitative evidence is not visible.

Read-first score

Read-first score 38.7, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 50.

Recency 5%

100

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

Topical relevance 25%

71.4

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

Methodology quality 15%

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

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: 461.

Keyword Scores

world model

interactive world model

video world model

generative world model

world simulator

world dynamics prediction

model-based reinforcement learning world model

Deep Analysis

Innovations

Hybrid Sparse Attention (HSA) that integrates linear global attention with head-adaptive sparse attention.
Hierarchical KV Cache (HKV) that organizes historical KV pairs into semantically indexed pages across multi-tier memory for fine-grained retrieval and controlled GPU residency.
Tailored attention kernels co-designed with the architecture to translate theoretical efficiency into real-world performance.

Methodology

WorldAttention is a system-oriented attention architecture for text-conditioned interactive video world models built on autoregressive diffusion. It combines Hybrid Sparse Attention and Hierarchical KV Cache, supported by specialized kernels. The approach is evaluated on VBench-Long and InterVBench against prior state-of-the-art methods using subject consistency as a reported metric.

Key Results

WorldAttention surpasses prior state-of-the-art methods, achieving subject consistency scores of 0.9472 on VBench-Long and 0.9668 on InterVBench.

Limitations

Sliding-window mechanisms inherently sacrifice historical context, undermining long-range interactive capabilities.
Maintaining a full-history cache remains computationally prohibitive and memory-intensive due to quadratic attention overhead and linear KV cache growth leading to GPU memory saturation.

Resources

Paper