The Year in AI Papers
The year/Independent research

Paper 2601.20540

Advancing Open-source World Models

Robbyant Team, Zelin Gao, Qiuyu Wang, Yanhong Zeng, Jiapeng Zhu, Ka Leong Cheng, Yixuan Li, Hanlin Wang, Yinghao Xu, Shuailei Ma, Yihang Chen, Jie Liu, et al.

Published
Jan 2026
Research lab
Independent
Citations
87
GitHub
4.3K stars

01 In brief

Summary

LingBot-World is an open-source world simulator built from video generation, offering high-fidelity dynamics across diverse environments, minute-level long-term memory, and real-time interactivity with sub-second latency.

It uses a three-stage training pipeline: pre-training on a 14B-parameter Wan2.2 model, middle-training with a mixture-of-experts (MoE) architecture (28B total) for action control and long-term consistency, and post-training with causal attention and few-step distillation for real-time inference.

The data engine combines general videos, game data with action signals, and synthetic Unreal Engine data, annotated with hierarchical captions.

Evaluations on VBench show superior dynamic degree (0.8857) and overall consistency compared to Yume-1.5 and HY-World 1.5.

Applications include promptable world events, action agents, and 3D reconstruction.

Limitations include memory instability, high computational cost, limited action space, and single-agent support.

The model is fully open-sourced with code, checkpoints, and weights available on GitHub and Hugging Face, aiming to democratize world model research and bridge the gap between open and closed-source technologies.

Future work focuses on expanding action space, improving memory stability, and enabling longer generation horizons.

02 From the paper

Abstract

We present LingBot-World, an open-sourced world simulator stemming from video generation. Positioned as a top-tier world model, LingBot-World offers the following features. (1) It maintains high fidelity and robust dynamics in a broad spectrum of environments, including realism, scientific contexts, cartoon styles, and beyond. (2) It enables a minute-level horizon while preserving contextual consistency over time, which is also known as "long-term memory". (3) It supports real-time interactivity, achieving a latency of under 1 second when producing 16 frames per second. We provide public access to the code and model in an effort to narrow the divide between open-source and closed-source technologies. We believe our release will empower the community with practical applications across areas like content creation, gaming, and robot learning.