Robo-bench — Arm
Agents · 2026-10-13
Mean ordered-milestone progress on Robo-bench's 13 arm tasks: a Franka Panda with a parallel-jaw gripper, wrist force-torque sensing, overhead and wrist cameras, joint position targets at up to 50 Hz and protective force stops. One three-minute real-time episode per task.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-6.1 Sol | 50.8 |
| GPT-6 Astra | 42.3 |
| Claude Opus 5.5 | 25.4 |
| Muse Wonder Full | 17.7 |
| Claude Fable 5.1 | 10.8 |
| Gemini 4 Argon | 1.5 |
| Grok 4.7 | 0.4 |