Alibaba unveils Qwen Robot Suite, its first AI models for physical robots

Alibaba's Tongyi Lab announced the Qwen Robot Suite, a trio of foundation models aimed at physical robots: Qwen-RobotNav for navigation, Qwen-RobotManip for object manipulation, and Qwen-RobotWorld, a world model that predicts future physical states. Reported by SCMP, the launch marks Alibaba's deliberate push beyond chatbots into embodied, agent-capable systems.

The three-model split mirrors the emerging structure of robotics stacks — perception/navigation, action/manipulation, and predictive world modeling — and signals Alibaba intends to compete across the full embodied-AI pipeline rather than a single component. The models are already in pilot testing with selected Alibaba Cloud enterprise clients, suggesting a commercial path through its cloud business rather than a pure research release.

The move places Alibaba alongside NVIDIA (whose Cosmos 3 physical-AI foundation model debuted earlier in June) and a wave of robotics-focused labs racing to bring large-model capabilities to machines. For China's AI ecosystem, it extends the Qwen brand — already strong in open-weight language models — into a strategically important domain where China has manufacturing and deployment advantages. Skeptics will want to see real-world manipulation benchmarks and generalization data, the hard parts of embodied AI, before judging how far ahead the suite actually is.