09 July 2026 | News
Image Courtesy: Public Domain
Robbyant, an embodied AI company within Ant Group, announced the open-source release of LingBot-World 2.0 (Infinity). This interactive world model significantly upgraded its world prediction and interactivity capabilities, supporting hour-long continuous generation, 720p/60fps high-definition real-time output, and richer interactive actions.
LingBot-World 2.0 also integrates a native agent mechanism, evolving generated worlds from merely watchable and controllable to "sustainably interactive and dynamically evolving", a first in the world model sector.
While LingBot-World 1.0 set a benchmark with minutes-level stable generation, version 2.0 pushes the boundaries toward true infinity. The model leverages a Causal Pretraining Paradigm combined with a proprietary MoBA (Mask of Bidirectional Attention) mechanism. This allows the model to learn world evolution in chronological order, effectively eliminating the compounding errors that typically cause texture blurring, geometry collapse, and scene breakdown in long-horizon generation. In rigorous hour-long stress tests, the model maintains visual fidelity with zero quality drift.
Alongside the open-source release of the pre-trained model, Robbyant has distilled a dedicated fast inference version and systematically optimized the generation pipeline. This is engineered to guarantee a seamless, gaming-like interactive experience for users.
Available now on the Reactor platform, it allows users to bypass the wait for complete sequence generation. Instead, LingBot-World 2.0 generates, transmits, and displays content simultaneously, delivering a stable output of 720p/60fps high-definition visuals, ensuring low latency feedback, and enabling users to control character movement or switch perspectives via keyboard in real time.
Beyond basic navigation, LingBot-World 2.0 dramatically expands the scope of interactivity.
l Rich Action Space: The model supports diverse character actions, such as attacking, shooting arrows, casting spells, jumping, and gliding. Outcomes are dynamically generated based on real-time scene states to maintain physical plausibility and visual consistency. Users can also trigger global events such as day-night cycles, weather changes, and entity injection via text commands.
l Dual-Agent Mechanism: The model features a built-in dual-agent mechanism. A Pilot Agent is responsible for planning and executing character behaviors, while a Director Agent dynamically introduces new events as the scene progresses.
l Multiplayer Persistence: The model supports multiple users within a single persistent world, enabling collaborative exploration and interaction, a critical step toward AI-native multiplayer experiences.
Committed to open innovation, Robbyant has released LingBot-World 2.0 under an open-source license with day-0 support for SGLang. Now, users and developers can experience LingBot-World 2.0 through the following channels:
Try Online (Reactor): https://reactor.inc/lingbot-world-v2
GitHub: https://github.com/Robbyant/lingbot-world-v2
Hugging Face: https://huggingface.co/collections/robbyant/lingbot-world-v2
Together with LingBot-World 2.0, Robbyant today also open-sourced LingBot-Video, the world’s first open-source video generation foundation model built on a Mixture-of-Experts (MoE) architecture specifically designed for embodied intelligence. Redesigned for robotics, it delivers significant improvements in inference efficiency, physical plausibility, action comprehension, and task completion, providing a new open-source foundation for transitioning video models from digital content creation to real-world embodied AI.
