RoboProcessBench - Motion Direction: leaderboard
Metric: Accuracy (%) on 1,207 questions asking the dominant motion direction of the manipulator, from a short clip (4 options, chance 25.0); 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 July 2027. 14 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Sonnet 4.6 | 54.5 |
| 2 | Qwen 3 VL 32B | 50.7 |
| 3 | GPT-5.4 Mini | 46.2 |
| 4 | GLM-4.6V | 42.7 |
| 5 | Qwen 2.5 VL 7B Instruct | 35.5 |
| 6 | InternVL3-38B | 33.2 |
| 7 | InternVL3-8B | 31.1 |
| 8 | InternVL3.5-8B | 27.7 |
| 9 | Claude Haiku 4.5 | 27.1 |
| 10 | GPT-4o | 26.1 |
Interactive version: theaggregate.ai/benchmark?slug=roboprocessbench-motion-direction · How It Works · Data refreshed daily, snapshot 2026-09-29.