The frontier of world models is shifting from passive video generation to fully interactive, persistent simulations. Robbyant, the embodied-intelligence arm of Ant Group, has released LingBot-World-Infinity (also known as LingBot-World 2.0), a 14B parameter causal video generation model that functions as a high-fidelity interactive world simulator.

Add AlexTech.ai asPreferred Source on Google

Solving the Long-Horizon Drift

Most interactive AI worlds suffer from "long-horizon drift," where textures smear and geometry warps as the simulation progresses. LingBot-World 2.0 tackles this through a novel Mixture of Bidirectional and Autoregressive (MoBA) attention mask.

蚂蚁灵波开源LingBot-World：国产世界模型实现10分钟无损生成与毫秒级交互 - 玉米小站 — https://www.yumiok.com/archives/4588.html

By pairing this architecture with distribution matching distillation over extensive self-rollout trajectories, the model achieves unprecedented stability. It can generate 720p output at 60 fps for over an hour without quality decay, effectively removing the temporal boundaries that previously limited AI-generated environments.

Agentic Interaction and Accessibility

Unlike traditional video tools, LingBot-World-Infinity is designed for active exploration. Users can navigate the generated spaces using standard controls, while an agentic harness allows built-in AI agents to trigger dynamic events in real time, creating a living virtual ecosystem.

To ensure broad accessibility, Robbyant has released both the primary 14B model and a streamlined 1.3B parameter version, the latter enabling deployment on a single consumer GPU. This open-source approach, supported by an Apache-2.0 license and availability on Hugging Face, positions the model as a formidable open alternative to Google Genie 3.

Implications for AI Training

The ability to generate unbounded, consistent synthetic environments marks a significant leap for agentic reinforcement learning. By providing an infinite stream of diverse, high-quality training data, LingBot-World 2.0 could drastically accelerate the development of embodied AI and autonomous robotics in safe, simulated spaces.