Ant's LingBot-Video: Open-Source MoE Video Model for Robotics

thetripathi58 · x · 2026-07-09

Ant Group's Robbyant has released LingBot-Video, the first large-scale open-source embodied intelligence video foundation model based on a MoE architecture. Designed specifically for robotics, it can function as a data engine, policy evaluator, and action planner, simulating future execution trajectories based on initial images and action sequences. The model utilizes a fully customized training approach, finding that multiple smaller expert models perform better than a few large ones, with scalability validated in the 120B version.

Related event: Ant Group Open-Sources LingBot-Video for Embodied AI(26 posts)→

Original post →

More from Embodied

Embodied channel →