SO-101 Bench leaderboard
Language-conditioned spatial manipulation benchmark for a single SO-101 arm, run in an Isaac Lab twin with a real-robot rank-transfer check. All learned policies share the same 3,629 sim demos and training budget; the authors run no public submission process.
Low-cost arm · SO-101 · Isaac Lab (digital twin)Report an error
83.2%Best · GPT-6.1 Sol (PyRUA) + Galaxea G0.5
11Models ranked
OpenSaturation
Nov 18, 2026Numbers checked
| # | Model | Type | Score | Date | Source |
|---|---|---|---|---|---|
| 1 | Agent system | 83.2%, self-reported by the authors, not a leaderboard entry | Oct 27, 2026 | ||
| 2 | Agent system | 66.8%, self-reported by the authors, not a leaderboard entry | Oct 27, 2026 | ||
| 3 | World-action | 61.4%, self-reported by the authors, not a leaderboard entry | Oct 27, 2026 | ||
| 4 | VLA | 60.8%, self-reported by the authors, not a leaderboard entry | Oct 27, 2026 | ||
| 5 | VLA | 45.8%, self-reported by the authors, not a leaderboard entry | Oct 27, 2026 | ||
| 6 | World-action | 43.0%, self-reported by the authors, not a leaderboard entry | Oct 27, 2026 | ||
| 7 | VLA | 41.0%, self-reported by the authors, not a leaderboard entry | Oct 27, 2026 | ||
| 8 | World-action | 39.9%, self-reported by the authors, not a leaderboard entry | Oct 27, 2026 | ||
| 9 | VLA | 23.8%, self-reported by the authors, not a leaderboard entry | Oct 27, 2026 | ||
| 10 | VLA | 20.9%, self-reported by the authors, not a leaderboard entry | Oct 27, 2026 | ||
| 11 | World-action | 8.0%, self-reported by the authors, not a leaderboard entry | Oct 27, 2026 |
Metric: Mean success over 15 task-family × object-split cells (916 episodes, 95% bootstrap CI). One result per model: RoboFYI re-run, then official leaderboard, then paper, then third party. Runs that deviate from the official protocol are listed on the model page but not ranked.
Wrong or missing number? Report an error