LingBot-Video: New MoE Video Foundation Model Released

@_akhaliq· July 8, 2026 View original

▶ The 2-minute explainer

Summary

LingBot-Video, a new Mixture-of-Experts (MoE) based video foundation model, has been released on Hugging Face, designed for embodied intelligence with 30 billion parameters and augmented with extensive embodied data.

A new video foundation model, LingBot-Video, is now available on Hugging Face. This model utilizes a Mixture-of-Experts (MoE) architecture, featuring 30 billion parameters, though only 3 billion are actively used during inference, optimizing for efficiency. LingBot-Video is specifically engineered for embodied intelligence applications. It has been significantly enhanced by pre-training on large-scale internet video data, further augmented with an additional 70,000 hours of embodied data, making it highly capable for tasks requiring understanding and interaction within physical or simulated environments.

Why it matters

This model offers advanced capabilities for developing AI in robotics, virtual agents, and interactive simulations, potentially accelerating progress in embodied AI.

How to implement this in your domain

  1. 1Access LingBot-Video on Hugging Face to explore its capabilities for embodied AI tasks.
  2. 2Integrate the model into robotic control systems for enhanced video understanding and decision-making.
  3. 3Utilize its MoE architecture for efficient inference in real-time embodied intelligence applications.
  4. 4Experiment with the model for generating realistic video responses in virtual agent interactions.
  5. 5Evaluate its performance on tasks requiring understanding of physical actions and environments.

Who benefits

RoboticsAI ResearchGamingVirtual RealityAutomotive

Key takeaways

  • LingBot-Video is an MoE-based video foundation model for embodied intelligence.
  • It features 30 billion parameters with efficient 3 billion active inference.
  • The model is trained on vast internet video and 70K hours of embodied data.
  • It holds potential for advancing robotics and interactive AI systems.

Original post by @_akhaliq

"LingBot-Video is out on Hugging Face MoE-based video foundation model built for embodied intelligence 30B params, only 3B active at inference Augmented with 70K hours of embodied data on top of large-scale internet video pretraining"

View on X

Originally posted by @_akhaliq on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses