AgiBot Open-Sources GO-1, Achieving 32% Performance Boost in General-Purpose Embodied AI

AgiBot has officially open-sourced its Genie Operator-1 (GO-1) model, a groundbreaking general-purpose embodied AI, on September 19, 2025. This model leverages the innovative Vision-Language-Latent-Action (ViLLA) architecture, designed for flexible and adaptable use across various robotic platforms. The announcement, shared by Tuo Liu, marks a significant step towards democratizing advanced robotics and enhancing the capabilities of embodied AI systems.

GO-1 introduces the novel ViLLA framework, which combines a Vision-Language Model (VLM) with a Mixture of Experts (MoE) to achieve superior generalization and dexterity. This hierarchical framework processes web-scale data, including human videos and robot demonstrations, through latent action representations. According to AgiBot, GO-1 significantly outperforms previous state-of-the-art models, demonstrating a 32% increase in success rates, from 46% to 78%, across complex tasks.

The foundation of GO-1's capabilities lies in the AgiBot World dataset, a large-scale, high-quality real-world robotics dataset. Comprising over 1 million trajectories across 217 tasks in five diverse application domains, AgiBot World provides the extensive and varied data necessary for training generalist policies. This dataset, along with the model, is now open-source, aiming to foster community-wide innovation in robotic manipulation.

AgiBot, a company valued at over $1 billion, has been at the forefront of developing advanced robotics, including humanoid robots like the Linxi X2. The open-sourcing of GO-1 and its underlying dataset reflects AgiBot's strategy to accelerate ecosystem adoption and drive progress in embodied AI. This move is expected to lower barriers for developers and researchers, paving the way for more robust and adaptable robotic systems in various real-world scenarios.

The release of GO-1 is poised to significantly impact the field of embodied AI by providing a powerful, adaptable model that can learn from diverse data and generalize across different tasks. While challenges in generalization beyond pre-trained scenarios remain an industry-wide hurdle, AgiBot's initiative aims to propel the development of more intelligent and versatile robots. This open-source approach seeks to empower the community to build upon this foundation, advancing the pursuit of scalable and general-purpose robotic intelligence.