114B Parameters, 6B Activated: Sand.ai Just Open-Sourced the World's First 100-Billion MoE Video Generation Model

Sand.ai has officially open-sourced a video generation model based on the Mixture of Experts (MoE) architecture. The model has a total of 114 billion parameters, but only activates 6 billion parameters during inference. This design aims to balance generation quality with computational resource consumption, representing a technical attempt in the field of video generation.

The core advantage of the model lies in its high efficiency-to-cost ratio, capable of generating 10-second 1080P resolution videos at a lower cost. The MoE architecture allows the model to invoke specific expert modules for specific tasks, avoiding the high computing power requirements brought by full parameter loading, making high-quality video generation more accessible.

As the world's first open-sourced 100-billion level MoE video generation model, this release lowers the R&D threshold in related fields. Developers can directly fine-tune or conduct secondary development based on this model, expected to accelerate the deployment of AI video applications in scenarios such as content creation, entertainment, and education.

This page provides an editorial summary based on publicly available information. It is not a republished article. Use the source link below for the original report.