AGIBOT unveils GO-3 Preview, robot hits 85.7% success in store trial

Chinese robotics firm AGIBOT introduced GO-3 Preview, an embodied AI model built on a World Reasoning Action Model architecture that links task reasoning, outcome prediction and action generation. It topped three embodied AI benchmarks: RoboDojo, RoboTwin and RoboColiseum. In a real convenience store, a G2 robot raised its full-workflow success rate from 28.6% on day one to 85.7% after five days; the full model is slated for open-source release by the end of 2026.

• GO-3 Preview ranked first on RoboDojo, RoboTwin and RoboColiseum
• G2 robot's success rate rose from 28.6% to 85.7% over five days
• After four hours of online practice the robot beat average human curling performance
• Full model planned for open-source release by end of 2026