LIBERO (VLA Manipulation) libero Leaderboard
LIBERO lifelong-robot-manipulation benchmark β mean success rate across the 4 standard suites (Spatial, Object, Goal, Long). Aggregated by the AllenAI vla-evaluation-harness leaderboard from published Vision-Language-Action papers. Β· Metric: Success Rate (higher is better)
| # | Model | Success Rate | Paper |
|---|---|---|---|
| 1 | LaST-R1 | 99.90 | link |
| 2 | QuoVLA | 99.60 | link |
| 3 | AcceRL | 99.40 | link |
| 4 | CORAL (SimVLA) | 99.30 | link |
| 5 | FUTURE-VLA | 99.20 | link |
| 6 | SimpleVLA-RL (from paper2603.18464) | 99.20 | link |
| 7 | SRPO (Online) | 99.20 | link |
| 8 | PriorVLA | 99.10 | link |
| 9 | GeoAlign | 99.00 | link |
| 10 | pi_0.5 with MoH | 99.00 | link |
| 11 | SimpleVLA-RL | 99.00 | link |
| 12 | DualCoT-VLA | 98.80 | link |
| 13 | 3DThinkVLA | 98.70 | link |
| 14 | Xiaomi-Robotics-0 | 98.70 | link |
| 15 | ABot-M0 | 98.60 | link |
| 16 | Cosmos Policy + World2Act | 98.60 | link |
| 17 | DiT4DiT | 98.60 | link |
| 18 | FocusVLA | 98.60 | link |
| 19 | Hume | 98.60 | link |
| 20 | LaWAM | 98.60 | link |
| 21 | Multi-view VLA + AML (from paper2605.11832) | 98.60 | link |
| 22 | SimVLA | 98.60 | link |
| 23 | Ο-EVA Stage 3 (from paper2606.09457) | 98.60 | link |
| 24 | Cosmos Policy | 98.50 | link |
| 25 | ElasticFlow | 98.50 | link |
| 26 | GeoSem-WAM | 98.50 | link |
| 27 | LingBot-VA | 98.50 | link |
| 28 | LingBot-VA (from Fast-WAM) | 98.50 | link |
| 29 | Spatial Forcing (from FocusVLA) | 98.50 | link |
| 30 | StreamVLA | 98.50 | link |