Robo-bench — Ninja Course (public)
Agents · 2026-10-13
Public Robo-bench humanoid task: climb stairs, ride a swinging platform across a pit, descend a ramp, crawl through a 0.8 m space and reach the finish, with milestones at 20, 35, 50, 60, 80 and 100. A single public, single-seed episode nested in the humanoid suite, so it is a diagnostic.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Fable 5.1 | 0.0 |
| Claude Opus 5.5 | 0.0 |
| Gemini 4 Argon | 0.0 |
| GPT-6.1 Sol | 0.0 |
| GPT-6 Astra | 0.0 |
| Grok 4.7 | 0.0 |
| Muse Wonder Full | 0.0 |