A Chinese artificial intelligence start-up has claimed a leading position in global robotics research, highlighting China’s growing momentum in so-called embodied intelligence.
Spirit AI said its latest foundation model, Spirit v1.5, ranked first on the RoboChallenge leaderboard, outperforming a leading US rival. The results were reported by Xinhua.
According to the benchmark, Spirit v1.5 achieved a total score of 66.09 and a task success rate of 50.33 per cent, making it the only evaluated model to cross the 50 per cent threshold. It finished ahead of the pi0.5 model developed by Physical Intelligence.
RoboChallenge is widely regarded in the industry as a rigorous “global exam” for robots, testing AI models across 30 real-world physical tasks, from object manipulation and tool use to target recognition in everyday environments.
Spirit AI said the performance reflects the strength of its Vision-Language-Action architecture, which integrates perception, reasoning and execution into a single end-to-end system, reducing errors common in traditional modular designs. In a move aimed at accelerating adoption and research, the company has open-sourced Spirit v1.5 and related resources.
The company is based in Hangzhou, an emerging hub for robotics and artificial intelligence that is also home to firms such as DeepSeek and Unitree Robotics. Spirit AI drew attention in mid-2025 with the launch of its Moz1 humanoid robot, designed for industrial and logistics use.
Industry specialists caution that embodied intelligence is still some distance from mass-market deployment. However, Spirit v1.5’s performance is seen as a significant step toward practical service robots, with the company forecasting a marked increase in capable robotic systems within the next two to three years.