AGIBOT published further details of GO-3 Preview on November 3, following its October 29 event, describing an embodied model that combines task understanding, future-state prediction and robot actions.
The company reports leading results on three robotics benchmarks and says it intends to release the model openly by the end of 2026. Those benchmark claims are AGIBOT's; we have not reproduced them. A planned release is not an available download, and its precise license and components should be checked when published.
The proposal matters because real robot work involves contact, changing surroundings and manipulation, not just responding to text. Real-world reliability and transfer across different machines remain questions for independent evaluation.
Reported from linked primary sources. No product or model was independently tested.
