LingBot-Video 30B MoE Video Model Released for Embodied AI
July 8, 2026
LingBot-Video is a 30B parameter Mixture-of-Experts video foundation model featuring 3B active parameters during inference. It was trained on large-scale internet video plus 70K hours of specialized embodied intelligence data.
HOW THIS AFFECTS YOU
●
builderThe MoE architecture allows for large model capabilities with significantly lower inference compute requirements.
●
researcherThe inclusion of 70K hours of embodied data offers a new benchmark for training models for physical interaction.