Figure AI has introduced Helix 2.5, an AI control model designed for its Figure 03 humanoid robot platform. The software manages perception, decision-making, and physical control. The model achieved a 56% success rate in performing household tasks across 30 previously unseen homes in the Bay Area, completing 237 out of 420 attempts.
The testing included three specific chores: making beds, folding towels, and tidying living rooms. The robot achieved a 67% success rate for making beds, 62% for folding towels, and 40% for tidying toys. These results represent a significant improvement over the 9% success rate baseline established by a model trained from scratch.
The system utilizes zero-shot generalization, meaning the robot performed tasks without prior training in the specific environments. This performance is supported by the Index dataset, a collection of human behavior recordings launched by Figure on August 25, 2026. Helix 2.5 builds upon the architecture of the previous Helix 02 model released in January 2026.
CEO Brett Adcock identified the Helix 2.5 project as the most significant undertaking for the company to date. While the 56% success rate marks a technical advancement in zero-shot generalization, the company acknowledges that the current reliability level does not yet meet the requirements for everyday household use.
