This article examines the performance of PI models π0 and π0.5 in RoboChallenge benchmarks, highlighting their high success rates in robotic tasks. It contrasts these models with the WALL-OSS-Flow's poor results, providing insights into current challenges in robotic foundational models. RoboChallenge's platform is portrayed as a key tool for objective evaluation of embodied AI systems, offering reproducible metrics and transparent comparison. The discussion targets researchers and developers in robotics and AI fields, aiming to identify reliable, high-performing models for practical applications.