RoboProcessBench - Primitive Chain Restoration: leaderboard

Metric: Accuracy (%) on 46 questions asking which primitive fills a masked slot in the local action chain (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 December 2026. 14 models tracked.

Top models

#ModelScore
1InternVL3-8B91.3
2GPT-5.4 Mini87
3InternVL3-38B82.6
4InternVL3.5-8B80.4
5Qwen 2.5 VL 7B Instruct80.4
6GLM-4.6V76.7
7GPT-4o76.1
8Qwen 3 VL 32B76.1
9Claude Haiku 4.567.4
10Claude Sonnet 4.665.2

Interactive version: theaggregate.ai/benchmark?slug=roboprocessbench-primitive-chain-restoration · How It Works · Data refreshed daily, snapshot 2026-09-29.