VLA Harness LIBERO: leaderboard
Language-conditioned robot manipulation across the LIBERO suites, the de facto evaluation for fine-tuned vision-language-action policies. Scores are the best figure any published paper reports per policy.
Metric: Success Rate (%). Source: allenai.github.io. 134 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | SimpleVLA-RL | 99.2 | |
| 2 | Being-H0.5 | 98.9 | |
| 3 | Faster-WAM | 98.9 | |
| 4 | Xiaomi-Robotics-0 | 98.7 | |
| 5 | SimVLA | 98.6 | |
| 6 | Cosmos Policy | 98.5 | |
| 7 | Spatial Forcing | 98.5 | |
| 8 | LingBot-VA | 98.5 | |
| 9 | MemoryVLA | 98.5 | |
| 10 | LangForce | 98.4 |
Interactive version: theaggregate.ai/benchmark?slug=vla-harness-libero · How It Works · Data refreshed daily, snapshot 2026-10-20.