Dual-Arm Sim Progress Score
Agents · 2026-10-30
Dual-Arm Sim's leaderboard metric over the same 80 trials per model: each trial earns credit for the furthest stage it reached (0 no contact, 20 approached, 40 grasped, 60 moved, 80 placed, 100 completed), averaged over the trials. Stages come from simulator ground truth, not from the model's report.
Top models (higher is better)
| Model | Score |
|---|---|
| Gemini 4 Argon | 49.5 |
| GPT-6 Astra | 49.5 |
| GPT-6.1 Sol | 40.8 |
| Gemini 3.8 Flash | 38.8 |
| Claude Fable 5.1 | 37.8 |
| Claude Sonnet 5.5 | 26.0 |
| Claude Opus 5.5 | 19.0 |
| Muse Wonder | 9.5 |
| Gemini 3.1 Pro Preview | 8.5 |
| Grok 4.7 | 6.3 |