VLA Harness RLBench: leaderboard
Multi-task robot manipulation in RLBench, spanning precise placement, articulated-object handling and tool use from demonstrations.
Metric: Success Rate (%). Source: allenai.github.io. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | BridgeVLA | 93.7 | |
| 2 | SAM2Act | 86.8 | |
| 3 | ARP | 86 | |
| 4 | RVT-2 | 81.4 | |
| 5 | 3D Diffuser Actor | 81.3 | |
| 6 | RVT2 | 79.3 | |
| 7 | Act3D | 65 | |
| 8 | RVT | 62.9 | |
| 9 | PerAct | 49.4 | |
| 10 | PolarNet | 46.4 |
Interactive version: theaggregate.ai/benchmark?slug=vla-harness-rlbench · How It Works · Data refreshed daily, snapshot 2026-10-20.