Natura Will has unveiled Natura-1, the first general-purpose foundation model for the physical world. This hybrid expert model, boasting 50 billion parameters, has been pre-trained on 1 million hours of first-person human experience data. Taking language and vision as inputs, the model can drive a dexterous hand with 20 degrees of freedom within a 100-millisecond control loop. Natura-1 is capable of following complex natural language instructions with zero-shot learning, demonstrating strong generalization capabilities in unfamiliar scenarios and tasks, and autonomously completing long, sequential workflows. It can also perform complex operations such as folding deformable materials, writing calligraphy, and switching tools in multi-stage tasks. The founder of Natura Will, Ding Ning, hails from the Department of Computer Science at Tsinghua University, and the team possesses strong scientific research capabilities. The company is committed to creating general physical intelligence and has independently developed a full-stack infrastructure covering data lakehouses, end-to-end data pipelines, underlying operators, and distributed training engines. Natura-1 is the first model trained based on this technology stack, with all system components designed to support further scaling.