SPEAR-1 is a robotics foundation AI model built by INSAIT that achieves state-of-the-art performance while using 20x less robotics data. A first version of SPEAR-1 was featured in WIRED back in October 2025.
The key breakthrough behind SPEAR-1 is a new pre-training approach which enables learning from robotics data, but also from non-robotics 3D data that is much more abundant and easier to obtain.
This ability to understand and locate objects in 3D space significantly boosts SPEAR-1’s robot manipulation capabilities and represents a meaningful step toward addressing one of robotics’ biggest challenges, the data bottleneck.
Our robotics team will be in Denver this week to present SPEAR-1 at CVPR 2026. If you are excited about building robotics foundation models, come find us on Sunday, 7 June, during poster session 5, at poster 606.
Learn more about SPEAR-1: https://spear.insait.ai/