RoboTwin-Clean2Rand
Can the model plan, use tools, and finish a multi-step task?
Agents4 ranked models5 reported values1 reportsHigher is better
Top rankings
One row per model, using its best reported score across effort settings.
| Rank | Model | Best score | Best reported setting |
|---|---|---|---|
| 1 | Qwen-RobotManipQwen | 82.4% | Reported configurationQwen report2 values · 1 reportJun 15, 2026 · source |
| 2 | π₀.₅Physical Intelligence | 67.0% | Reported configurationQwen report1 value · 1 reportJun 15, 2026 · source |
| 3 | GR00T-N1.7NVIDIA | 40.4% | Reported configurationQwen report1 value · 1 reportJun 15, 2026 · source |
| 4 | StarVLAExternal baseline | 27.1% | Reported configurationQwen report1 value · 1 reportJun 15, 2026 · source |
Effort curve
Every sourced cost-linked effort value for this exact version. Lines connect complete sweeps only.
No cost-linked effort sweep for this version.
Definition and comparison boundarypublic methodology
Can the model plan, use tools, and finish a multi-step task?
Agentic execution across tools, environments, or long-running workflows. Higher is better. The value is the percentage reported in this lab's table.
Harness, tools, prompts, attempt count, and benchmark version can materially change the result. Compare within one reporting context.
Clean-set fine-tuning followed by progressive environmental randomization; values are success rates.
Method / source