1X Technologies has integrated its new video-pretrained world model, 1XWM, into the NEO robot platform. This model targets robotics researchers and developers focused on advancing home robots capable of human-like navigation and interaction. Currently, the release is limited to a select group for research and evaluation, with plans for broader deployment after further validation. The 1XWM model shifts away from traditional vision-language-action (VLA) frameworks, utilizing data from internet-scale videos and egocentric human and robot interactions. It predicts robot actions by generating text-conditioned video rollouts, which are then converted into motion commands through an Inverse Dynamics Model. This approach significantly reduces the need for extensive robot demonstration hours, allowing for quicker task adaptation. The model is built on a 14 billion parameter generative video framework, optimized for NEO's humanoid design, with each rollout taking about 11 seconds. In early tests, 1XWM shows better generalization to new objects and actions compared to models from other labs. It effectively handles complex tasks like bimanual coordination and object manipulation, achieving success rates that meet or surpass previous models. Feedback indicates that the use of egocentric human data and detailed captions during training enhances the robot's reliability and physical plausibility. For this launch, 1X collaborated with Verda to improve inference speed, aiming to reduce latency and broaden the model's potential for household autonomy.