LingBot-Video Performance and MoE Inference Acceleration

thetripathi58 · x · 2026-07-09

In the public RBench test, LingBot-Video scored an average of 0.620, leading other open-source models in manipulation, long-horizon, and quadruped tasks. However, the team admits it still lags behind the closed-source Wan 2.6 in spatial and robotic arm tasks.

Thanks to its MoE architecture, the model has 30B total parameters but only activates 3B per inference step. When processing 1 million tokens of long video, it runs 3.18 times faster than standard 30B dense models, enabling large model execution at extremely low cost.

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

Original post →

More from Embodied

Embodied channel →