Particle.news
Download on the App Store

1X Releases Video World Model Enabling Prompt-to-Action Learning on NEO

The company says the update learns from internet-scale human video to generalize skills, shifting improvement toward data gathered through real-world deployment.

Overview

  • 1X introduced a video-based World Model for its NEO humanoid that converts short voice or text prompts into physical actions in situations the robot has not seen before, leveraging a humanlike design for transfer from human video.
  • Demonstrations show NEO packing a lunch box and performing previously unseen tasks such as opening a sliding door, operating a toilet seat, ironing clothes, and brushing a person’s hair.
  • 1X positions progress as deployment-led, saying capability will scale with the number of robots collecting real-world experience rather than with human-operated training data.
  • The company outlines layered safety that combines compliant, low-energy hardware with AI alignment designed to reason about risks and choose safer action paths.
  • Early access is available via 1X’s online store at $20,000 with a $499-per-month option and deliveries planned for 2026, while TechCrunch reports the system reflects an ongoing training loop rather than instant one-shot mastery.