AgiBot Launches GO-3 Preview Embodied AI Model
The new model leverages world reasoning architecture to top robotic performance leaderboards.
Updated on Oct. 30, 2026 in Robotics
Live Poll
Do you trust that integrating AI robots into everyday services will improve efficiency and quality?
AgiBot has introduced its GO-3 Preview embodied AI model, which uses a World Reasoning Action Model architecture to integrate task reasoning with physical motion. The model recently secured the top position on the RoboDojo, RoboTwin, and RoboColiseum leaderboards.
Why it matters
By combining task reasoning, future outcome prediction, and action generation, the model significantly enhances how robots perform physical tasks. This advancement demonstrates a major shift in how embodied systems learn to interact with complex environments.
The Genie Master framework improved single-task success by 17.7 percentage points compared to conventional supervised fine-tuning. Additionally, robots achieved this performance jump after just five days of training, with systems surpassing human performance in four hours.
The players
AgiBot
A robotics technology company that specializes in the development of embodied AI and hardware systems for physical task automation.
The details
The system utilizes the Genie Evolver distributed training architecture, which bridges the gap between real-world robots and physics simulations. In a test case involving G2 robots within a convenience store environment, the framework successfully scaled task flow completion rates from 28.6 percent to 85.7 percent.
Timeline
AgiBot launched the GO-3 Preview model on October 29, 2026.
AgiBot plans to fully open-source the GO-3 model by the end of 2026.
The Tech Race
This release continues the industry-wide transition toward embodied AI systems that can reason about future outcomes rather than merely reacting to visual inputs. It positions AgiBot to challenge existing robotic automation standards that rely on static, supervised fine-tuning.
The increased efficiency in these robots could lead to faster and more reliable automated services in retail and convenience store settings. Users may soon interact with hardware that is more capable of handling complex, multi-step physical tasks in real-time.
The takeaway
The rapid improvement from 28.6 percent to 85.7 percent success rates highlights the growing maturity of embodied AI training systems. Integrating physical robots with virtual world models is proving to be a highly effective way to accelerate task learning.
What happens next
AgiBot has committed to fully open-sourcing the GO-3 model by the end of 2026.
Further reading
For more information on the current state of autonomous systems, visit the Robotics section.
Live Poll
Do you trust that integrating AI robots into everyday services will improve efficiency and quality?
