CITIC Securities: General-purpose models entering the embodied task domain will become the next investment direction for the AI industry.

CITIC Securities research report notes that general-purpose large models are extending from high-level task planning for robots further into motion generation and execution control. The robot brain mainly relies on specialized models such as VLA and WAM trained on robot motion data to translate visual and language instructions into actions. Recently, research institutions including NVIDIA and Stanford have successively explored applying GPT-6 Astra to robot control, attempting to leverage the general model's existing visual understanding, spatial reasoning, and task planning capabilities to reduce reliance on specialized policy training. Astra has demonstrated good generalization on embodied task datasets. We believe that looking ahead to 2027, general models entering the embodied task domain will become the AI industry's next investment direction, while in the short term general models and specialized models are more likely to form a division of labor and collaboration.