Figure's Helix 2.5 does chores in 30 unseen homes

Figure introduced Helix 2.5 on September 17, reporting that a single foundation model pretrained on its Index human-video dataset let Figure 03 tidy living rooms, fold towels and make beds across 30 Bay Area homes with no data collected, fine-tuning or adaptation in any of them. Holding task data and architecture fixed, Index pretraining lifted whole-task zero-shot success from 9% to 56%, or 237 of 420 trials with no partial credit, and Figure says the same behaviours needed half the task-specific data of a comparable Helix 02 policy. The company also reports a human-to-robot transfer scaling law precise enough to forecast its largest run's loss before training, says Index now generates about 35 minutes of new human experience every second, and has committed $3.5 billion of compute to Helix.