VLA Harness LIBERO-Plus: leaderboard
A robustness-hardened LIBERO variant that perturbs objects, camera viewpoints and language to test whether reported manipulation gains survive distribution shift.
Metric: Success Rate (%). Source: allenai.github.io. 36 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | pi-0.5 | 86.5 | |
| 2 | π0.5 | 85.6 | |
| 3 | VLA-Adapter | 85.3 | |
| 4 | π_0.5 | 85.2 | |
| 5 | pi-0.₅ | 85.1 | |
| 6 | VLANeXt | 84.5 | |
| 7 | Cosmos-Policy | 83.1 | |
| 8 | ImageWAM | 82.9 | |
| 9 | MemoryVLA | 81.5 | |
| 10 | OpenVLA-OFT | 80.7 |
Interactive version: theaggregate.ai/benchmark?slug=vla-harness-libero-plus · How It Works · Data refreshed daily, snapshot 2026-10-20.