← Leaderboard

Helix 2.5

VLA
Figure AI · Sep 17, 2026

Helix variant pretrained from random initialization on Index, Figure’s human-behavior dataset, instead of starting from a VLM. Results are self-reported: in blind zero-shot tests across 30 unseen homes (tidying, towel folding, bed making), the Index-pretrained policy reached 56% success against 9% for one trained from scratch.

Type
VLA
Parameters
—
Backbone
–
Weights
Closed
License
–
Commercial use
Unclear
Released
Sep 17, 2026
Robots
Figure humanoid
SO-101
–
LeRobot
–
Inputs
–
VRAM
–

Benchmark results

0 of 13 benchmarks

No benchmark results recorded yet.

Not reported on 13 benchmarks
Wrong or missing number? Report an error