Act2Answer - Symmetry: leaderboard

Metric: Soft success rate (%; 150 binary-choice items from IconQA, 300 episodes; answer by placing a cube on the chosen image tile in a Simpler tabletop episode, original and left/right-swapped layouts, chance 50). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 6 models tracked.

Top models

#ModelScore
1Xiaomi-Robotics-R058
2π046
3OpenVLA45
4SpatialVLA45
5InternVLA-M143
6Magma-8B37

Interactive version: theaggregate.ai/benchmark?slug=act2answer-symmetry · How It Works · Data refreshed daily, snapshot 2026-09-25.