Ant Group's Robbyant Unveils LingBot-World-Infinity for Dynamic AI Simulation
Ant Group's Robbyant unit has released LingBot-World-Infinity, an open-source 14-billion-parameter model that redefines interactive world simulation with hour-long, stable video generation and intelligent agent interaction.
✨ This content was summarized and interpreted by AI; it may contain errors — please verify accuracy with the original sources. Learn more
Listen to this story

Robbyant, Ant Group's embodied-intelligence unit, has unveiled LingBot-World-Infinity (LingBot-World 2.0), an open-source 14-billion-parameter causal video generation model that dramatically pushes the boundaries of interactive world simulation. This release marks a significant leap, enabling hour-long continuous generation at 720p and 60 frames per second without the visual drift or structural degradation that plagued previous models.
The model’s core innovation lies in its "Mixture of Bidirectional and Autoregressive (MoBA) Attention Mask," which acts as a regularizer to prevent overfitting and maintain visual fidelity over extended interactions, a key challenge in long-form video generation. LingBot-World-Infinity also features an "agentic harness" that pairs a Pilot Agent for character behavior with a Director Agent for synthesizing dynamic environmental events and ensuring causal reasoning. This mechanism transforms simulated worlds from passive backdrops into "sustainably interactive and dynamically evolving" environments, supporting a broad action space including combat, spell-casting, and real-time environmental changes.
This development signals a critical shift in AI, moving beyond mere content generation to creating truly intelligent agents capable of understanding and interacting with complex, dynamic environments. By focusing on causal pretraining, LingBot-World-Infinity learns how actions fundamentally alter environments, which is crucial for advancements in robotics, virtual training, and future-state prediction. While challenges remain in achieving true long-term memory and perfectly faithful physical understanding, the open-source nature of LingBot-World-Infinity promises to accelerate research, laying a robust foundation for AI systems that genuinely comprehend and adapt to the physical world. This innovation positions Ant Group at the forefront of developing AI that not only generates content but truly understands and interacts with its simulated reality, paving the way for more sophisticated and adaptive AI applications across various industries.