Hour-Long Interactive Worlds
The release marks a substantial leap from the original LingBot-World, which was open-sourced in January and supported sequences of roughly 10 minutes. LingBot-World 2.0, subtitled “Infinity,” upgrades both world prediction and interactivity, allowing users to navigate and act inside generated environments in real time. The model is available on GitHub and Hugging Face under an open-source license with day-zero support for SGLang, a serving framework for large generative models. afp.com marktechpost.com lasvegassun.com
LingBot-Video: A MoE Foundation Model
Alongside the world model, Robbyant also open-sourced LingBot-Video, which it describes as the first open-source video generation foundation model built on a Mixture-of-Experts architecture designed specifically for embodied intelligence. The model uses a 13-billion-parameter total capacity with 1.4 billion active parameters per token, offering improved inference efficiency and physical plausibility for robotics applications. Weights, code, and integration with both Diffusers and SGLang are included in the release. reddit.com lasvegassun.com afp.com
A Growing Open-Source Stack
The twin releases cap a prolific stretch for Robbyant, which has open-sourced a series of embodied-AI components in 2026 including LingBot-Map for streaming 3D reconstruction in April and LingBot-Depth 2.0 for spatial perception earlier this month. The Next Web noted that the pace amounts to “a full embodied-AI stack” released in a single week. Developers can try LingBot-World 2.0 through an online demo on Reactor or clone the repository directly from GitHub. thenextweb.com morningstar.com scmp.com lasvegassun.com
