88Z AI, 9 July — Ant Lingbo Technology has open-sourced LingBot-Video, the world's first Mixture-of-Experts (MoE) foundation model for embodied AI video generation. With 30 billion total parameters and only 3 billion activated during inference, it offers triple the efficiency of dense architectures. The model leverages a 70,000-hour dataset and multi-dimensional reinforcement learning rewards to improve physical plausibility and task completion.
At a glance
- LingBot-Video is the world's first MoE-based video generation model for embodied AI, offering three times the inference efficiency of dense architectures.
- The model leverages a 70,000-hour embodied dataset and multi-dimensional reinforcement learning rewards to enhance physical plausibility and task completion.
The update
Ant Lingbo Technology has officially open-sourced LingBot-Video, the world's first video generation foundation model designed specifically for embodied intelligence, utilizing an advanced Mixture-of-Experts (MoE) architecture. The model features 30 billion total parameters but activates only about 3 billion during inference, achieving three times the efficiency of comparable dense architecture models and significantly reducing computational costs. To enhance performance in real-world physical environments, the research team constructed a proprietary embodied AI dataset spanning 70,000 hours. Furthermore, the model incorporates a multi-dimensional reinforcement learning reward system that focuses on optimizing physical plausibility and task completion, ensuring generated content aligns better with realistic logic and operational requirements.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- 蚂蚁灵波科技Source published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.