Ant Group Open-Sources LingBot-World 2.0 Interactive World Model

Robbyant, an embodied AI company under Ant Group, released LingBot-World 2.0 (also known as LingBot-World-Infinity), an open-source interactive causal world model. Moving beyond traditional video clip generation, this model pushes AI towards sustainably interactive "game worlds," drawing significant attention for its hour-level stable generation and multiplayer capabilities.

Core Tech and Performance

Utilizing a causal pre-training and post-training approach, the model specifically tackles the common autoregressive drift issue in world models (where accumulating errors distort the generated world). It achieved 720p/60fps real-time rendering over a continuous 60-minute rollout across 20 scenes, with almost no noticeable visual degradation. It requires only a single starting frame, combined with real-time camera movement or text input, to generate 3D environments.

Gameplay and Application Potential

Hailed as the first AI video model that plays like a real video game, LingBot-World 2.0 allows players to control characters using WASD keys, cast spells, and summon weather. Its interaction system uses a background VLM to suggest real-time events bound to keyboard hotkeys, categorized into fixed keys and context-aware events. Furthermore, it supports multiplayer, allowing multiple players to explore the same procedurally generated AI world simultaneously, likened to an AI version of Minecraft. Beyond entertainment, it holds potential for embodied simulations.

Open Source Availability

The project is fully open, with the author releasing model weights, GitHub code, an online demo, demonstration videos, and related papers.

2026-07-11 ~ 2026-07-12 · 15 related posts

Full story(2 episodes)→

2 near-duplicate retellings: kimmonismus · LearnWithBishal