RoboProcessBench - Motion State: leaderboard
RoboProcessBench (2026) asks vision-language models multiple-choice questions about how a robot manipulation is unfolding, using frames and short clips from held-out recordings of four robot datasets. This task asks whether the manipulator is actively moving or effectively stationary, with 2 options; accuracy in percent, higher is better.
Metric: Accuracy (%) on 896 questions asking whether the manipulator is actively moving or effectively stationary (2 options, chance 50.0); zero-shot multiple-choice VQA on held-out robot manipulation recordings from GM-100, RH20T, REASSEMBLE and AIST-Bimanual (ProcessData-Eval), temperature 0.01; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o | 68.4 | #741 |
| 2 | Claude Sonnet 4.6 | 68.1 | #161 |
| 3 | GPT-5.4 Mini | 67 | #389 |
| 4 | Claude Haiku 4.5 | 61.3 | #568 |
| 5 | InternVL3-38B | 58.1 | #903 |
| 6 | Qwen 2.5 VL 7B Instruct | 58 | #1833 |
| 7 | InternVL3-8B | 54 | #1688 |
| 8 | GLM-4.6V | 48.9 | #701 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=roboprocessbench-motion-state · How It Works · Data refreshed daily, snapshot 2026-11-13.