Act2Answer - Living World: leaderboard
Act2Answer (Kachaev, Moskalenko et al., 2026) turns two-option knowledge questions into short Simpler tabletop episodes in which a vision-language-action policy answers by placing a cube on one of two image tiles; each item runs in its original and left/right-swapped layout. This category covers animal and plant identity and living-versus-nonliving distinctions, adapted from open-ended OK-VQA by pairing each image with a same-category distractor: 150 items, 300 episodes. The score is the percentage of episodes ending on the correct tile (chance 50); higher is better.
Metric: Soft success rate (%; 150 binary-choice items from OK-VQA, 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
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Magma-8B | 77 | |
| 2 | InternVLA-M1 | 58 | |
| 3 | Xiaomi-Robotics-R0 | 56 | |
| 4 | OpenVLA | 52 | |
| 5 | SpatialVLA | 49 | |
| 6 | π0 | 45 |
Interactive version: theaggregate.ai/benchmark?slug=act2answer-living-world · How It Works · Data refreshed daily, snapshot 2026-11-12.