Qiyuan Robotics has launched PrimeSim, a world-model simulator designed for robot training that creates interactive, trial-and-error-capable, and continuously learning virtual environments to support policy validation and reinforcement learning. The model has topped the WorldSimProbe LIBERO leaderboard, ranking first in three of five tests on the comprehensive LIBERO leaderboard. When used for reinforcement learning, it can boost policy success rates from 61.7% to 73.2%.