Alibaba releases Qwen-RobotSuite, three embodied-AI models for the physical world

Category: AI
Status: developing
Sources: Qwen blog, Qwen-VLA repository, discussion
Summary: Alibaba launched Qwen-RobotSuite on 2026-06-16: Qwen-RobotManip, a vision-language-action manipulation model built on Qwen3.5-4B; Qwen-RobotNav, a vision-language navigation model built on Qwen3-VL at 2B, 4B, and 8B sizes; and Qwen-RobotWorld, a language-conditioned video world model using a 60-layer MMDiT with a frozen Qwen2.5-VL encoder. RobotManip and RobotNav ship with public GitHub repositories. Alibaba reports RobotManip was trained on more than 38,000 hours of open-source data and topped the RoboChallenge generalist track with a process score of 59.83 and a 45 percent task success rate.
Why it matters: A frontier open-weight lab moves its vision-language stack into manipulation and navigation, extending the open-weight model race into embodied robotics.
Follow-up: Track independent reproduction of the RoboChallenge numbers, license terms, and which weights are released.