RoboProcessBench - Current Primitive Recognition: leaderboard
Metric: Accuracy (%) on 359 questions asking which low-level primitive is being executed, 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 2035. 14 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | InternVL3.5-8B | 36.8 |
| 2 | GLM-4.6V | 34 |
| 3 | Qwen 2.5 VL 7B Instruct | 33.1 |
| 4 | GPT-5.4 Mini | 32.1 |
| 5 | GPT-4o | 32 |
| 6 | InternVL3-8B | 31.2 |
| 7 | Qwen 3 VL 32B | 28.1 |
| 8 | Claude Sonnet 4.6 | 27.3 |
| 9 | InternVL3-38B | 26.2 |
| 10 | Claude Haiku 4.5 | 23.4 |
Interactive version: theaggregate.ai/benchmark?slug=roboprocessbench-current-primitive-recognition · How It Works · Data refreshed daily, snapshot 2026-09-29.