Matrix-Game 3.5: Enhancing Real-Time Streaming Interactive World Models with Patch Memory

arXiv 2026 38 method, system

TLDR

Matrix-Game 3.5 improves real-time interactive world generation with geometry-aware patch memory, static-dynamic disentanglement, and progressive distillation for long-horizon consistency.

Reasoning

The paper presents well-motivated architectural innovations for interactive world modeling, including memory, disentanglement, and distillation. However, the visible abstract provides no explicit experimental results or benchmarks, and it ends mid-sentence, limiting verification of claimed real-world performance.

Read-first score

Read-first score 38, 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%

10.9

PageRank-style prestige over the local citation graph built from available OpenAlex references. local_pagerank=0.0031

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

Frontier

Rank sensitivity

Stability: volatile; rank range: 400.

Keyword Scores

world model

interactive world model

generative world model

world simulator

video world model

world dynamics prediction

model-based reinforcement learning world model

Deep Analysis

Innovations

Unified geometry-aware memory framework combining patch-memory and tiled-PRoPE for explicit 3D patch retrieval and projective camera conditioning without additional learnable parameters
Static-dynamic disentangled world representation that separately models static scene geometry and dynamic subjects to preserve geometric consistency and subject identity
Two-stage progressive real-time distillation framework using Perceptual Flow Matching and curriculum-based Self-Rollout DMD to convert a bidirectional diffusion model into a few-step causal generator

Methodology

The paper builds on Matrix-Game 3.0 and introduces a geometry-aware memory framework with patch-memory and tiled-PRoPE, a static-dynamic disentangled world representation, and a two-stage progressive distillation framework. The model is trained on a unified corpus spanning Unreal simulation environments, open-world games, and internet videos. The distillation process converts a bidirectional diffusion model into a few-step causal generator for real-time autoregressive generation.

Key Results

Experiments demonstrate strong performance in long-horizon scene recall, precise camera control, subject consistency, prompt-driven world generation, and stable real-time open-world interaction, supporting minute-long real-time interactive generation.

Resources

Paper