ATOM-Bench - Single-Arm Instruction: leaderboard
The Instruction Set of ATOM-Bench (Wu et al., 2026) on a real Franka Panda arm: seven pick-and-place tasks that each isolate one language constraint (color, shape, size, spatial relation to a reference object, exact count, destination, exclusion). Success rate in percent over 10 fixed physical test seeds per task after fine-tuning on 100 human demonstrations per atomic task, macro-averaged; higher is better.
Metric: Success rate (%; Instruction Set, 7 tasks x 10 fixed physical test seeds on the Franka Panda single-arm track, macro-averaged over tasks; one policy fine-tuned on the 15 atomic tasks of the track). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Pi0.5 | 94.3 | |
| 2 | Motus | 67.1 | |
| 3 | GR00T-N1.6 | 57.1 | |
| 4 | LingBot-VLA | 54.3 | |
| 5 | SmolVLA | 11.4 |
Interactive version: theaggregate.ai/benchmark?slug=atom-bench-single-arm-instruction · How It Works · Data refreshed daily, snapshot 2026-11-12.